Commit b73f186a authored by antirez's avatar antirez
Browse files

Modules documentation removed from source.

Moving to redis-doc repository to publish via Redis.io.
parent 09d93ec9
# Modules API reference
## `RM_Alloc`
void *RM_Alloc(size_t bytes);
Use like malloc(). Memory allocated with this function is reported in
Redis INFO memory, used for keys eviction according to maxmemory settings
and in general is taken into account as memory allocated by Redis.
You should avoid using malloc().
## `RM_Calloc`
void *RM_Calloc(size_t nmemb, size_t size);
Use like calloc(). Memory allocated with this function is reported in
Redis INFO memory, used for keys eviction according to maxmemory settings
and in general is taken into account as memory allocated by Redis.
You should avoid using calloc() directly.
## `RM_Realloc`
void* RM_Realloc(void *ptr, size_t bytes);
Use like realloc() for memory obtained with `RedisModule_Alloc()`.
## `RM_Free`
void RM_Free(void *ptr);
Use like free() for memory obtained by `RedisModule_Alloc()` and
`RedisModule_Realloc()`. However you should never try to free with
`RedisModule_Free()` memory allocated with malloc() inside your module.
## `RM_Strdup`
char *RM_Strdup(const char *str);
Like strdup() but returns memory allocated with `RedisModule_Alloc()`.
## `RM_PoolAlloc`
void *RM_PoolAlloc(RedisModuleCtx *ctx, size_t bytes);
Return heap allocated memory that will be freed automatically when the
module callback function returns. Mostly suitable for small allocations
that are short living and must be released when the callback returns
anyway. The returned memory is aligned to the architecture word size
if at least word size bytes are requested, otherwise it is just
aligned to the next power of two, so for example a 3 bytes request is
4 bytes aligned while a 2 bytes request is 2 bytes aligned.
There is no realloc style function since when this is needed to use the
pool allocator is not a good idea.
The function returns NULL if `bytes` is 0.
## `RM_GetApi`
int RM_GetApi(const char *funcname, void **targetPtrPtr);
Lookup the requested module API and store the function pointer into the
target pointer. The function returns `REDISMODULE_ERR` if there is no such
named API, otherwise `REDISMODULE_OK`.
This function is not meant to be used by modules developer, it is only
used implicitly by including redismodule.h.
## `RM_IsKeysPositionRequest`
int RM_IsKeysPositionRequest(RedisModuleCtx *ctx);
Return non-zero if a module command, that was declared with the
flag "getkeys-api", is called in a special way to get the keys positions
and not to get executed. Otherwise zero is returned.
## `RM_KeyAtPos`
void RM_KeyAtPos(RedisModuleCtx *ctx, int pos);
When a module command is called in order to obtain the position of
keys, since it was flagged as "getkeys-api" during the registration,
the command implementation checks for this special call using the
`RedisModule_IsKeysPositionRequest()` API and uses this function in
order to report keys, like in the following example:
if (`RedisModule_IsKeysPositionRequest(ctx))` {
`RedisModule_KeyAtPos(ctx`,1);
`RedisModule_KeyAtPos(ctx`,2);
}
Note: in the example below the get keys API would not be needed since
keys are at fixed positions. This interface is only used for commands
with a more complex structure.
## `RM_CreateCommand`
int RM_CreateCommand(RedisModuleCtx *ctx, const char *name, RedisModuleCmdFunc cmdfunc, const char *strflags, int firstkey, int lastkey, int keystep);
Register a new command in the Redis server, that will be handled by
calling the function pointer 'func' using the RedisModule calling
convention. The function returns `REDISMODULE_ERR` if the specified command
name is already busy or a set of invalid flags were passed, otherwise
`REDISMODULE_OK` is returned and the new command is registered.
This function must be called during the initialization of the module
inside the `RedisModule_OnLoad()` function. Calling this function outside
of the initialization function is not defined.
The command function type is the following:
int MyCommand_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv, int argc);
And is supposed to always return `REDISMODULE_OK`.
The set of flags 'strflags' specify the behavior of the command, and should
be passed as a C string compoesd of space separated words, like for
example "write deny-oom". The set of flags are:
* **"write"**: The command may modify the data set (it may also read
from it).
* **"readonly"**: The command returns data from keys but never writes.
* **"admin"**: The command is an administrative command (may change
replication or perform similar tasks).
* **"deny-oom"**: The command may use additional memory and should be
denied during out of memory conditions.
* **"deny-script"**: Don't allow this command in Lua scripts.
* **"allow-loading"**: Allow this command while the server is loading data.
Only commands not interacting with the data set
should be allowed to run in this mode. If not sure
don't use this flag.
* **"pubsub"**: The command publishes things on Pub/Sub channels.
* **"random"**: The command may have different outputs even starting
from the same input arguments and key values.
* **"allow-stale"**: The command is allowed to run on slaves that don't
serve stale data. Don't use if you don't know what
this means.
* **"no-monitor"**: Don't propoagate the command on monitor. Use this if
the command has sensible data among the arguments.
* **"fast"**: The command time complexity is not greater
than O(log(N)) where N is the size of the collection or
anything else representing the normal scalability
issue with the command.
* **"getkeys-api"**: The command implements the interface to return
the arguments that are keys. Used when start/stop/step
is not enough because of the command syntax.
* **"no-cluster"**: The command should not register in Redis Cluster
since is not designed to work with it because, for
example, is unable to report the position of the
keys, programmatically creates key names, or any
other reason.
## `RM_SetModuleAttribs`
void RM_SetModuleAttribs(RedisModuleCtx *ctx, const char *name, int ver, int apiver);
Called by `RM_Init()` to setup the `ctx->module` structure.
This is an internal function, Redis modules developers don't need
to use it.
## `RM_Milliseconds`
long long RM_Milliseconds(void);
Return the current UNIX time in milliseconds.
## `RM_AutoMemory`
void RM_AutoMemory(RedisModuleCtx *ctx);
Enable automatic memory management. See API.md for more information.
The function must be called as the first function of a command implementation
that wants to use automatic memory.
## `RM_CreateString`
RedisModuleString *RM_CreateString(RedisModuleCtx *ctx, const char *ptr, size_t len);
Create a new module string object. The returned string must be freed
with `RedisModule_FreeString()`, unless automatic memory is enabled.
The string is created by copying the `len` bytes starting
at `ptr`. No reference is retained to the passed buffer.
## `RM_CreateStringPrintf`
RedisModuleString *RM_CreateStringPrintf(RedisModuleCtx *ctx, const char *fmt, ...);
Create a new module string object from a printf format and arguments.
The returned string must be freed with `RedisModule_FreeString()`, unless
automatic memory is enabled.
The string is created using the sds formatter function sdscatvprintf().
## `RM_CreateStringFromLongLong`
RedisModuleString *RM_CreateStringFromLongLong(RedisModuleCtx *ctx, long long ll);
Like `RedisModule_CreatString()`, but creates a string starting from a long long
integer instead of taking a buffer and its length.
The returned string must be released with `RedisModule_FreeString()` or by
enabling automatic memory management.
## `RM_CreateStringFromString`
RedisModuleString *RM_CreateStringFromString(RedisModuleCtx *ctx, const RedisModuleString *str);
Like `RedisModule_CreatString()`, but creates a string starting from another
RedisModuleString.
The returned string must be released with `RedisModule_FreeString()` or by
enabling automatic memory management.
## `RM_FreeString`
void RM_FreeString(RedisModuleCtx *ctx, RedisModuleString *str);
Free a module string object obtained with one of the Redis modules API calls
that return new string objects.
It is possible to call this function even when automatic memory management
is enabled. In that case the string will be released ASAP and removed
from the pool of string to release at the end.
## `RM_RetainString`
void RM_RetainString(RedisModuleCtx *ctx, RedisModuleString *str);
Every call to this function, will make the string 'str' requiring
an additional call to `RedisModule_FreeString()` in order to really
free the string. Note that the automatic freeing of the string obtained
enabling modules automatic memory management counts for one
`RedisModule_FreeString()` call (it is just executed automatically).
Normally you want to call this function when, at the same time
the following conditions are true:
1) You have automatic memory management enabled.
2) You want to create string objects.
3) Those string objects you create need to live *after* the callback
function(for example a command implementation) creating them returns.
Usually you want this in order to store the created string object
into your own data structure, for example when implementing a new data
type.
Note that when memory management is turned off, you don't need
any call to RetainString() since creating a string will always result
into a string that lives after the callback function returns, if
no FreeString() call is performed.
## `RM_StringPtrLen`
const char *RM_StringPtrLen(const RedisModuleString *str, size_t *len);
Given a string module object, this function returns the string pointer
and length of the string. The returned pointer and length should only
be used for read only accesses and never modified.
## `RM_StringToLongLong`
int RM_StringToLongLong(const RedisModuleString *str, long long *ll);
Convert the string into a long long integer, storing it at `*ll`.
Returns `REDISMODULE_OK` on success. If the string can't be parsed
as a valid, strict long long (no spaces before/after), `REDISMODULE_ERR`
is returned.
## `RM_StringToDouble`
int RM_StringToDouble(const RedisModuleString *str, double *d);
Convert the string into a double, storing it at `*d`.
Returns `REDISMODULE_OK` on success or `REDISMODULE_ERR` if the string is
not a valid string representation of a double value.
## `RM_StringCompare`
int RM_StringCompare(RedisModuleString *a, RedisModuleString *b);
Compare two string objects, returning -1, 0 or 1 respectively if
a < b, a == b, a > b. Strings are compared byte by byte as two
binary blobs without any encoding care / collation attempt.
## `RM_StringAppendBuffer`
int RM_StringAppendBuffer(RedisModuleCtx *ctx, RedisModuleString *str, const char *buf, size_t len);
Append the specified buffere to the string 'str'. The string must be a
string created by the user that is referenced only a single time, otherwise
`REDISMODULE_ERR` is returend and the operation is not performed.
## `RM_WrongArity`
int RM_WrongArity(RedisModuleCtx *ctx);
Send an error about the number of arguments given to the command,
citing the command name in the error message.
Example:
if (argc != 3) return `RedisModule_WrongArity(ctx)`;
## `RM_ReplyWithLongLong`
int RM_ReplyWithLongLong(RedisModuleCtx *ctx, long long ll);
Send an integer reply to the client, with the specified long long value.
The function always returns `REDISMODULE_OK`.
## `RM_ReplyWithError`
int RM_ReplyWithError(RedisModuleCtx *ctx, const char *err);
Reply with the error 'err'.
Note that 'err' must contain all the error, including
the initial error code. The function only provides the initial "-", so
the usage is, for example:
`RM_ReplyWithError(ctx`,"ERR Wrong Type");
and not just:
`RM_ReplyWithError(ctx`,"Wrong Type");
The function always returns `REDISMODULE_OK`.
## `RM_ReplyWithSimpleString`
int RM_ReplyWithSimpleString(RedisModuleCtx *ctx, const char *msg);
Reply with a simple string (+... \r\n in RESP protocol). This replies
are suitable only when sending a small non-binary string with small
overhead, like "OK" or similar replies.
The function always returns `REDISMODULE_OK`.
## `RM_ReplyWithArray`
int RM_ReplyWithArray(RedisModuleCtx *ctx, long len);
Reply with an array type of 'len' elements. However 'len' other calls
to `ReplyWith*` style functions must follow in order to emit the elements
of the array.
When producing arrays with a number of element that is not known beforehand
the function can be called with the special count
`REDISMODULE_POSTPONED_ARRAY_LEN`, and the actual number of elements can be
later set with `RedisModule_ReplySetArrayLength()` (which will set the
latest "open" count if there are multiple ones).
The function always returns `REDISMODULE_OK`.
## `RM_ReplySetArrayLength`
void RM_ReplySetArrayLength(RedisModuleCtx *ctx, long len);
When `RedisModule_ReplyWithArray()` is used with the argument
`REDISMODULE_POSTPONED_ARRAY_LEN`, because we don't know beforehand the number
of items we are going to output as elements of the array, this function
will take care to set the array length.
Since it is possible to have multiple array replies pending with unknown
length, this function guarantees to always set the latest array length
that was created in a postponed way.
For example in order to output an array like [1,[10,20,30]] we
could write:
`RedisModule_ReplyWithArray(ctx`,`REDISMODULE_POSTPONED_ARRAY_LEN`);
`RedisModule_ReplyWithLongLong(ctx`,1);
`RedisModule_ReplyWithArray(ctx`,`REDISMODULE_POSTPONED_ARRAY_LEN`);
`RedisModule_ReplyWithLongLong(ctx`,10);
`RedisModule_ReplyWithLongLong(ctx`,20);
`RedisModule_ReplyWithLongLong(ctx`,30);
`RedisModule_ReplySetArrayLength(ctx`,3); // Set len of 10,20,30 array.
`RedisModule_ReplySetArrayLength(ctx`,2); // Set len of top array
Note that in the above example there is no reason to postpone the array
length, since we produce a fixed number of elements, but in the practice
the code may use an interator or other ways of creating the output so
that is not easy to calculate in advance the number of elements.
## `RM_ReplyWithStringBuffer`
int RM_ReplyWithStringBuffer(RedisModuleCtx *ctx, const char *buf, size_t len);
Reply with a bulk string, taking in input a C buffer pointer and length.
The function always returns `REDISMODULE_OK`.
## `RM_ReplyWithString`
int RM_ReplyWithString(RedisModuleCtx *ctx, RedisModuleString *str);
Reply with a bulk string, taking in input a RedisModuleString object.
The function always returns `REDISMODULE_OK`.
## `RM_ReplyWithNull`
int RM_ReplyWithNull(RedisModuleCtx *ctx);
Reply to the client with a NULL. In the RESP protocol a NULL is encoded
as the string "$-1\r\n".
The function always returns `REDISMODULE_OK`.
## `RM_ReplyWithCallReply`
int RM_ReplyWithCallReply(RedisModuleCtx *ctx, RedisModuleCallReply *reply);
Reply exactly what a Redis command returned us with `RedisModule_Call()`.
This function is useful when we use `RedisModule_Call()` in order to
execute some command, as we want to reply to the client exactly the
same reply we obtained by the command.
The function always returns `REDISMODULE_OK`.
## `RM_ReplyWithDouble`
int RM_ReplyWithDouble(RedisModuleCtx *ctx, double d);
Send a string reply obtained converting the double 'd' into a bulk string.
This function is basically equivalent to converting a double into
a string into a C buffer, and then calling the function
`RedisModule_ReplyWithStringBuffer()` with the buffer and length.
The function always returns `REDISMODULE_OK`.
## `RM_Replicate`
int RM_Replicate(RedisModuleCtx *ctx, const char *cmdname, const char *fmt, ...);
Replicate the specified command and arguments to slaves and AOF, as effect
of execution of the calling command implementation.
The replicated commands are always wrapped into the MULTI/EXEC that
contains all the commands replicated in a given module command
execution. However the commands replicated with `RedisModule_Call()`
are the first items, the ones replicated with `RedisModule_Replicate()`
will all follow before the EXEC.
Modules should try to use one interface or the other.
This command follows exactly the same interface of `RedisModule_Call()`,
so a set of format specifiers must be passed, followed by arguments
matching the provided format specifiers.
Please refer to `RedisModule_Call()` for more information.
The command returns `REDISMODULE_ERR` if the format specifiers are invalid
or the command name does not belong to a known command.
## `RM_ReplicateVerbatim`
int RM_ReplicateVerbatim(RedisModuleCtx *ctx);
This function will replicate the command exactly as it was invoked
by the client. Note that this function will not wrap the command into
a MULTI/EXEC stanza, so it should not be mixed with other replication
commands.
Basically this form of replication is useful when you want to propagate
the command to the slaves and AOF file exactly as it was called, since
the command can just be re-executed to deterministically re-create the
new state starting from the old one.
The function always returns `REDISMODULE_OK`.
## `RM_GetClientId`
unsigned long long RM_GetClientId(RedisModuleCtx *ctx);
Return the ID of the current client calling the currently active module
command. The returned ID has a few guarantees:
1. The ID is different for each different client, so if the same client
executes a module command multiple times, it can be recognized as
having the same ID, otherwise the ID will be different.
2. The ID increases monotonically. Clients connecting to the server later
are guaranteed to get IDs greater than any past ID previously seen.
Valid IDs are from 1 to 2^64-1. If 0 is returned it means there is no way
to fetch the ID in the context the function was currently called.
## `RM_GetSelectedDb`
int RM_GetSelectedDb(RedisModuleCtx *ctx);
Return the currently selected DB.
## `RM_SelectDb`
int RM_SelectDb(RedisModuleCtx *ctx, int newid);
Change the currently selected DB. Returns an error if the id
is out of range.
Note that the client will retain the currently selected DB even after
the Redis command implemented by the module calling this function
returns.
If the module command wishes to change something in a different DB and
returns back to the original one, it should call `RedisModule_GetSelectedDb()`
before in order to restore the old DB number before returning.
## `RM_OpenKey`
void *RM_OpenKey(RedisModuleCtx *ctx, robj *keyname, int mode);
Return an handle representing a Redis key, so that it is possible
to call other APIs with the key handle as argument to perform
operations on the key.
The return value is the handle repesenting the key, that must be
closed with `RM_CloseKey()`.
If the key does not exist and WRITE mode is requested, the handle
is still returned, since it is possible to perform operations on
a yet not existing key (that will be created, for example, after
a list push operation). If the mode is just READ instead, and the
key does not exist, NULL is returned. However it is still safe to
call `RedisModule_CloseKey()` and `RedisModule_KeyType()` on a NULL
value.
## `RM_CloseKey`
void RM_CloseKey(RedisModuleKey *key);
Close a key handle.
## `RM_KeyType`
int RM_KeyType(RedisModuleKey *key);
Return the type of the key. If the key pointer is NULL then
`REDISMODULE_KEYTYPE_EMPTY` is returned.
## `RM_ValueLength`
size_t RM_ValueLength(RedisModuleKey *key);
Return the length of the value associated with the key.
For strings this is the length of the string. For all the other types
is the number of elements (just counting keys for hashes).
If the key pointer is NULL or the key is empty, zero is returned.
## `RM_DeleteKey`
int RM_DeleteKey(RedisModuleKey *key);
If the key is open for writing, remove it, and setup the key to
accept new writes as an empty key (that will be created on demand).
On success `REDISMODULE_OK` is returned. If the key is not open for
writing `REDISMODULE_ERR` is returned.
## `RM_GetExpire`
mstime_t RM_GetExpire(RedisModuleKey *key);
Return the key expire value, as milliseconds of remaining TTL.
If no TTL is associated with the key or if the key is empty,
`REDISMODULE_NO_EXPIRE` is returned.
## `RM_SetExpire`
int RM_SetExpire(RedisModuleKey *key, mstime_t expire);
Set a new expire for the key. If the special expire
`REDISMODULE_NO_EXPIRE` is set, the expire is cancelled if there was
one (the same as the PERSIST command).
Note that the expire must be provided as a positive integer representing
the number of milliseconds of TTL the key should have.
The function returns `REDISMODULE_OK` on success or `REDISMODULE_ERR` if
the key was not open for writing or is an empty key.
## `RM_StringSet`
int RM_StringSet(RedisModuleKey *key, RedisModuleString *str);
If the key is open for writing, set the specified string 'str' as the
value of the key, deleting the old value if any.
On success `REDISMODULE_OK` is returned. If the key is not open for
writing or there is an active iterator, `REDISMODULE_ERR` is returned.
## `RM_StringDMA`
char *RM_StringDMA(RedisModuleKey *key, size_t *len, int mode);
Prepare the key associated string value for DMA access, and returns
a pointer and size (by reference), that the user can use to read or
modify the string in-place accessing it directly via pointer.
The 'mode' is composed by bitwise OR-ing the following flags:
`REDISMODULE_READ` -- Read access
`REDISMODULE_WRITE` -- Write access
If the DMA is not requested for writing, the pointer returned should
only be accessed in a read-only fashion.
On error (wrong type) NULL is returned.
DMA access rules:
1. No other key writing function should be called since the moment
the pointer is obtained, for all the time we want to use DMA access
to read or modify the string.
2. Each time `RM_StringTruncate()` is called, to continue with the DMA
access, `RM_StringDMA()` should be called again to re-obtain
a new pointer and length.
3. If the returned pointer is not NULL, but the length is zero, no
byte can be touched (the string is empty, or the key itself is empty)
so a `RM_StringTruncate()` call should be used if there is to enlarge
the string, and later call StringDMA() again to get the pointer.
## `RM_StringTruncate`
int RM_StringTruncate(RedisModuleKey *key, size_t newlen);
If the string is open for writing and is of string type, resize it, padding
with zero bytes if the new length is greater than the old one.
After this call, `RM_StringDMA()` must be called again to continue
DMA access with the new pointer.
The function returns `REDISMODULE_OK` on success, and `REDISMODULE_ERR` on
error, that is, the key is not open for writing, is not a string
or resizing for more than 512 MB is requested.
If the key is empty, a string key is created with the new string value
unless the new length value requested is zero.
## `RM_ListPush`
int RM_ListPush(RedisModuleKey *key, int where, RedisModuleString *ele);
Push an element into a list, on head or tail depending on 'where' argumnet.
If the key pointer is about an empty key opened for writing, the key
is created. On error (key opened for read-only operations or of the wrong
type) `REDISMODULE_ERR` is returned, otherwise `REDISMODULE_OK` is returned.
## `RM_ListPop`
RedisModuleString *RM_ListPop(RedisModuleKey *key, int where);
Pop an element from the list, and returns it as a module string object
that the user should be free with `RM_FreeString()` or by enabling
automatic memory. 'where' specifies if the element should be popped from
head or tail. The command returns NULL if:
1) The list is empty.
2) The key was not open for writing.
3) The key is not a list.
## `RM_ZsetAddFlagsToCoreFlags`
int RM_ZsetAddFlagsToCoreFlags(int flags);
Conversion from/to public flags of the Modules API and our private flags,
so that we have everything decoupled.
## `RM_ZsetAddFlagsFromCoreFlags`
int RM_ZsetAddFlagsFromCoreFlags(int flags);
See previous function comment.
## `RM_ZsetAdd`
int RM_ZsetAdd(RedisModuleKey *key, double score, RedisModuleString *ele, int *flagsptr);
Add a new element into a sorted set, with the specified 'score'.
If the element already exists, the score is updated.
A new sorted set is created at value if the key is an empty open key
setup for writing.
Additional flags can be passed to the function via a pointer, the flags
are both used to receive input and to communicate state when the function
returns. 'flagsptr' can be NULL if no special flags are used.
The input flags are:
`REDISMODULE_ZADD_XX`: Element must already exist. Do nothing otherwise.
`REDISMODULE_ZADD_NX`: Element must not exist. Do nothing otherwise.
The output flags are:
`REDISMODULE_ZADD_ADDED`: The new element was added to the sorted set.
`REDISMODULE_ZADD_UPDATED`: The score of the element was updated.
`REDISMODULE_ZADD_NOP`: No operation was performed because XX or NX flags.
On success the function returns `REDISMODULE_OK`. On the following errors
`REDISMODULE_ERR` is returned:
* The key was not opened for writing.
* The key is of the wrong type.
* 'score' double value is not a number (NaN).
## `RM_ZsetIncrby`
int RM_ZsetIncrby(RedisModuleKey *key, double score, RedisModuleString *ele, int *flagsptr, double *newscore);
This function works exactly like `RM_ZsetAdd()`, but instead of setting
a new score, the score of the existing element is incremented, or if the
element does not already exist, it is added assuming the old score was
zero.
The input and output flags, and the return value, have the same exact
meaning, with the only difference that this function will return
`REDISMODULE_ERR` even when 'score' is a valid double number, but adding it
to the existing score resuts into a NaN (not a number) condition.
This function has an additional field 'newscore', if not NULL is filled
with the new score of the element after the increment, if no error
is returned.
## `RM_ZsetRem`
int RM_ZsetRem(RedisModuleKey *key, RedisModuleString *ele, int *deleted);
Remove the specified element from the sorted set.
The function returns `REDISMODULE_OK` on success, and `REDISMODULE_ERR`
on one of the following conditions:
* The key was not opened for writing.
* The key is of the wrong type.
The return value does NOT indicate the fact the element was really
removed (since it existed) or not, just if the function was executed
with success.
In order to know if the element was removed, the additional argument
'deleted' must be passed, that populates the integer by reference
setting it to 1 or 0 depending on the outcome of the operation.
The 'deleted' argument can be NULL if the caller is not interested
to know if the element was really removed.
Empty keys will be handled correctly by doing nothing.
## `RM_ZsetScore`
int RM_ZsetScore(RedisModuleKey *key, RedisModuleString *ele, double *score);
On success retrieve the double score associated at the sorted set element
'ele' and returns `REDISMODULE_OK`. Otherwise `REDISMODULE_ERR` is returned
to signal one of the following conditions:
* There is no such element 'ele' in the sorted set.
* The key is not a sorted set.
* The key is an open empty key.
## `RM_ZsetRangeStop`
void RM_ZsetRangeStop(RedisModuleKey *key);
Stop a sorted set iteration.
## `RM_ZsetRangeEndReached`
int RM_ZsetRangeEndReached(RedisModuleKey *key);
Return the "End of range" flag value to signal the end of the iteration.
## `RM_ZsetFirstInScoreRange`
int RM_ZsetFirstInScoreRange(RedisModuleKey *key, double min, double max, int minex, int maxex);
Setup a sorted set iterator seeking the first element in the specified
range. Returns `REDISMODULE_OK` if the iterator was correctly initialized
otherwise `REDISMODULE_ERR` is returned in the following conditions:
1. The value stored at key is not a sorted set or the key is empty.
The range is specified according to the two double values 'min' and 'max'.
Both can be infinite using the following two macros:
`REDISMODULE_POSITIVE_INFINITE` for positive infinite value
`REDISMODULE_NEGATIVE_INFINITE` for negative infinite value
'minex' and 'maxex' parameters, if true, respectively setup a range
where the min and max value are exclusive (not included) instead of
inclusive.
## `RM_ZsetLastInScoreRange`
int RM_ZsetLastInScoreRange(RedisModuleKey *key, double min, double max, int minex, int maxex);
Exactly like `RedisModule_ZsetFirstInScoreRange()` but the last element of
the range is selected for the start of the iteration instead.
## `RM_ZsetFirstInLexRange`
int RM_ZsetFirstInLexRange(RedisModuleKey *key, RedisModuleString *min, RedisModuleString *max);
Setup a sorted set iterator seeking the first element in the specified
lexicographical range. Returns `REDISMODULE_OK` if the iterator was correctly
initialized otherwise `REDISMODULE_ERR` is returned in the
following conditions:
1. The value stored at key is not a sorted set or the key is empty.
2. The lexicographical range 'min' and 'max' format is invalid.
'min' and 'max' should be provided as two RedisModuleString objects
in the same format as the parameters passed to the ZRANGEBYLEX command.
The function does not take ownership of the objects, so they can be released
ASAP after the iterator is setup.
## `RM_ZsetLastInLexRange`
int RM_ZsetLastInLexRange(RedisModuleKey *key, RedisModuleString *min, RedisModuleString *max);
Exactly like `RedisModule_ZsetFirstInLexRange()` but the last element of
the range is selected for the start of the iteration instead.
## `RM_ZsetRangeCurrentElement`
RedisModuleString *RM_ZsetRangeCurrentElement(RedisModuleKey *key, double *score);
Return the current sorted set element of an active sorted set iterator
or NULL if the range specified in the iterator does not include any
element.
## `RM_ZsetRangeNext`
int RM_ZsetRangeNext(RedisModuleKey *key);
Go to the next element of the sorted set iterator. Returns 1 if there was
a next element, 0 if we are already at the latest element or the range
does not include any item at all.
## `RM_ZsetRangePrev`
int RM_ZsetRangePrev(RedisModuleKey *key);
Go to the previous element of the sorted set iterator. Returns 1 if there was
a previous element, 0 if we are already at the first element or the range
does not include any item at all.
## `RM_HashSet`
int RM_HashSet(RedisModuleKey *key, int flags, ...);
Set the field of the specified hash field to the specified value.
If the key is an empty key open for writing, it is created with an empty
hash value, in order to set the specified field.
The function is variadic and the user must specify pairs of field
names and values, both as RedisModuleString pointers (unless the
CFIELD option is set, see later).
Example to set the hash argv[1] to the value argv[2]:
`RedisModule_HashSet(key`,`REDISMODULE_HASH_NONE`,argv[1],argv[2],NULL);
The function can also be used in order to delete fields (if they exist)
by setting them to the specified value of `REDISMODULE_HASH_DELETE`:
`RedisModule_HashSet(key`,`REDISMODULE_HASH_NONE`,argv[1],
`REDISMODULE_HASH_DELETE`,NULL);
The behavior of the command changes with the specified flags, that can be
set to `REDISMODULE_HASH_NONE` if no special behavior is needed.
`REDISMODULE_HASH_NX`: The operation is performed only if the field was not
already existing in the hash.
`REDISMODULE_HASH_XX`: The operation is performed only if the field was
already existing, so that a new value could be
associated to an existing filed, but no new fields
are created.
`REDISMODULE_HASH_CFIELDS`: The field names passed are null terminated C
strings instead of RedisModuleString objects.
Unless NX is specified, the command overwrites the old field value with
the new one.
When using `REDISMODULE_HASH_CFIELDS`, field names are reported using
normal C strings, so for example to delete the field "foo" the following
code can be used:
`RedisModule_HashSet(key`,`REDISMODULE_HASH_CFIELDS`,"foo",
`REDISMODULE_HASH_DELETE`,NULL);
Return value:
The number of fields updated (that may be less than the number of fields
specified because of the XX or NX options).
In the following case the return value is always zero:
* The key was not open for writing.
* The key was associated with a non Hash value.
## `RM_HashGet`
int RM_HashGet(RedisModuleKey *key, int flags, ...);
Get fields from an hash value. This function is called using a variable
number of arguments, alternating a field name (as a StringRedisModule
pointer) with a pointer to a StringRedisModule pointer, that is set to the
value of the field if the field exist, or NULL if the field did not exist.
At the end of the field/value-ptr pairs, NULL must be specified as last
argument to signal the end of the arguments in the variadic function.
This is an example usage:
RedisModuleString *first, *second;
`RedisModule_HashGet(mykey`,`REDISMODULE_HASH_NONE`,argv[1],&first,
argv[2],&second,NULL);
As with `RedisModule_HashSet()` the behavior of the command can be specified
passing flags different than `REDISMODULE_HASH_NONE`:
`REDISMODULE_HASH_CFIELD`: field names as null terminated C strings.
`REDISMODULE_HASH_EXISTS`: instead of setting the value of the field
expecting a RedisModuleString pointer to pointer, the function just
reports if the field esists or not and expects an integer pointer
as the second element of each pair.
Example of `REDISMODULE_HASH_CFIELD`:
RedisModuleString *username, *hashedpass;
`RedisModule_HashGet(mykey`,"username",&username,"hp",&hashedpass, NULL);
Example of `REDISMODULE_HASH_EXISTS`:
int exists;
`RedisModule_HashGet(mykey`,argv[1],&exists,NULL);
The function returns `REDISMODULE_OK` on success and `REDISMODULE_ERR` if
the key is not an hash value.
Memory management:
The returned RedisModuleString objects should be released with
`RedisModule_FreeString()`, or by enabling automatic memory management.
## `RM_FreeCallReply_Rec`
void RM_FreeCallReply_Rec(RedisModuleCallReply *reply, int freenested);
Free a Call reply and all the nested replies it contains if it's an
array.
## `RM_FreeCallReply`
void RM_FreeCallReply(RedisModuleCallReply *reply);
Wrapper for the recursive free reply function. This is needed in order
to have the first level function to return on nested replies, but only
if called by the module API.
## `RM_CallReplyType`
int RM_CallReplyType(RedisModuleCallReply *reply);
Return the reply type.
## `RM_CallReplyLength`
size_t RM_CallReplyLength(RedisModuleCallReply *reply);
Return the reply type length, where applicable.
## `RM_CallReplyArrayElement`
RedisModuleCallReply *RM_CallReplyArrayElement(RedisModuleCallReply *reply, size_t idx);
Return the 'idx'-th nested call reply element of an array reply, or NULL
if the reply type is wrong or the index is out of range.
## `RM_CallReplyInteger`
long long RM_CallReplyInteger(RedisModuleCallReply *reply);
Return the long long of an integer reply.
## `RM_CallReplyStringPtr`
const char *RM_CallReplyStringPtr(RedisModuleCallReply *reply, size_t *len);
Return the pointer and length of a string or error reply.
## `RM_CreateStringFromCallReply`
RedisModuleString *RM_CreateStringFromCallReply(RedisModuleCallReply *reply);
Return a new string object from a call reply of type string, error or
integer. Otherwise (wrong reply type) return NULL.
## `RM_Call`
RedisModuleCallReply *RM_Call(RedisModuleCtx *ctx, const char *cmdname, const char *fmt, ...);
Exported API to call any Redis command from modules.
On success a RedisModuleCallReply object is returned, otherwise
NULL is returned and errno is set to the following values:
EINVAL: command non existing, wrong arity, wrong format specifier.
EPERM: operation in Cluster instance with key in non local slot.
## `RM_CallReplyProto`
const char *RM_CallReplyProto(RedisModuleCallReply *reply, size_t *len);
Return a pointer, and a length, to the protocol returned by the command
that returned the reply object.
## `RM_CreateDataType`
moduleType *RM_CreateDataType(RedisModuleCtx *ctx, const char *name, int encver, void *typemethods_ptr);
Register a new data type exported by the module. The parameters are the
following. Please for in depth documentation check the modules API
documentation, especially the TYPES.md file.
* **name**: A 9 characters data type name that MUST be unique in the Redis
Modules ecosystem. Be creative... and there will be no collisions. Use
the charset A-Z a-z 9-0, plus the two "-_" characters. A good
idea is to use, for example `<typename>-<vendor>`. For example
"tree-AntZ" may mean "Tree data structure by @antirez". To use both
lower case and upper case letters helps in order to prevent collisions.
* **encver**: Encoding version, which is, the version of the serialization
that a module used in order to persist data. As long as the "name"
matches, the RDB loading will be dispatched to the type callbacks
whatever 'encver' is used, however the module can understand if
the encoding it must load are of an older version of the module.
For example the module "tree-AntZ" initially used encver=0. Later
after an upgrade, it started to serialize data in a different format
and to register the type with encver=1. However this module may
still load old data produced by an older version if the rdb_load
callback is able to check the encver value and act accordingly.
The encver must be a positive value between 0 and 1023.
* **typemethods_ptr** is a pointer to a RedisModuleTypeMethods structure
that should be populated with the methods callbacks and structure
version, like in the following example:
RedisModuleTypeMethods tm = {
.version = `REDISMODULE_TYPE_METHOD_VERSION`,
.rdb_load = myType_RDBLoadCallBack,
.rdb_save = myType_RDBSaveCallBack,
.aof_rewrite = myType_AOFRewriteCallBack,
.free = myType_FreeCallBack,
// Optional fields
.digest = myType_DigestCallBack,
.mem_usage = myType_MemUsageCallBack,
}
* **rdb_load**: A callback function pointer that loads data from RDB files.
* **rdb_save**: A callback function pointer that saves data to RDB files.
* **aof_rewrite**: A callback function pointer that rewrites data as commands.
* **digest**: A callback function pointer that is used for `DEBUG DIGEST`.
* **mem_usage**: A callback function pointer that is used for `MEMORY`.
* **free**: A callback function pointer that can free a type value.
The **digest* and **mem_usage** methods should currently be omitted since
they are not yet implemented inside the Redis modules core.
Note: the module name "AAAAAAAAA" is reserved and produces an error, it
happens to be pretty lame as well.
If there is already a module registering a type with the same name,
and if the module name or encver is invalid, NULL is returned.
Otherwise the new type is registered into Redis, and a reference of
type RedisModuleType is returned: the caller of the function should store
this reference into a gobal variable to make future use of it in the
modules type API, since a single module may register multiple types.
Example code fragment:
static RedisModuleType *BalancedTreeType;
int `RedisModule_OnLoad(RedisModuleCtx` *ctx) {
// some code here ...
BalancedTreeType = `RM_CreateDataType(`...);
}
## `RM_ModuleTypeSetValue`
int RM_ModuleTypeSetValue(RedisModuleKey *key, moduleType *mt, void *value);
If the key is open for writing, set the specified module type object
as the value of the key, deleting the old value if any.
On success `REDISMODULE_OK` is returned. If the key is not open for
writing or there is an active iterator, `REDISMODULE_ERR` is returned.
## `RM_ModuleTypeGetType`
moduleType *RM_ModuleTypeGetType(RedisModuleKey *key);
Assuming `RedisModule_KeyType()` returned `REDISMODULE_KEYTYPE_MODULE` on
the key, returns the moduel type pointer of the value stored at key.
If the key is NULL, is not associated with a module type, or is empty,
then NULL is returned instead.
## `RM_ModuleTypeGetValue`
void *RM_ModuleTypeGetValue(RedisModuleKey *key);
Assuming `RedisModule_KeyType()` returned `REDISMODULE_KEYTYPE_MODULE` on
the key, returns the module type low-level value stored at key, as
it was set by the user via `RedisModule_ModuleTypeSet()`.
If the key is NULL, is not associated with a module type, or is empty,
then NULL is returned instead.
## `RM_SaveUnsigned`
void RM_SaveUnsigned(RedisModuleIO *io, uint64_t value);
Save an unsigned 64 bit value into the RDB file. This function should only
be called in the context of the rdb_save method of modules implementing new
data types.
## `RM_LoadUnsigned`
uint64_t RM_LoadUnsigned(RedisModuleIO *io);
Load an unsigned 64 bit value from the RDB file. This function should only
be called in the context of the rdb_load method of modules implementing
new data types.
## `RM_SaveSigned`
void RM_SaveSigned(RedisModuleIO *io, int64_t value);
Like `RedisModule_SaveUnsigned()` but for signed 64 bit values.
## `RM_LoadSigned`
int64_t RM_LoadSigned(RedisModuleIO *io);
Like `RedisModule_LoadUnsigned()` but for signed 64 bit values.
## `RM_SaveString`
void RM_SaveString(RedisModuleIO *io, RedisModuleString *s);
In the context of the rdb_save method of a module type, saves a
string into the RDB file taking as input a RedisModuleString.
The string can be later loaded with `RedisModule_LoadString()` or
other Load family functions expecting a serialized string inside
the RDB file.
## `RM_SaveStringBuffer`
void RM_SaveStringBuffer(RedisModuleIO *io, const char *str, size_t len);
Like `RedisModule_SaveString()` but takes a raw C pointer and length
as input.
## `RM_LoadString`
RedisModuleString *RM_LoadString(RedisModuleIO *io);
In the context of the rdb_load method of a module data type, loads a string
from the RDB file, that was previously saved with `RedisModule_SaveString()`
functions family.
The returned string is a newly allocated RedisModuleString object, and
the user should at some point free it with a call to `RedisModule_FreeString()`.
If the data structure does not store strings as RedisModuleString objects,
the similar function `RedisModule_LoadStringBuffer()` could be used instead.
## `RM_LoadStringBuffer`
char *RM_LoadStringBuffer(RedisModuleIO *io, size_t *lenptr);
Like `RedisModule_LoadString()` but returns an heap allocated string that
was allocated with `RedisModule_Alloc()`, and can be resized or freed with
`RedisModule_Realloc()` or `RedisModule_Free()`.
The size of the string is stored at '*lenptr' if not NULL.
The returned string is not automatically NULL termianted, it is loaded
exactly as it was stored inisde the RDB file.
## `RM_SaveDouble`
void RM_SaveDouble(RedisModuleIO *io, double value);
In the context of the rdb_save method of a module data type, saves a double
value to the RDB file. The double can be a valid number, a NaN or infinity.
It is possible to load back the value with `RedisModule_LoadDouble()`.
## `RM_LoadDouble`
double RM_LoadDouble(RedisModuleIO *io);
In the context of the rdb_save method of a module data type, loads back the
double value saved by `RedisModule_SaveDouble()`.
## `RM_SaveFloat`
void RM_SaveFloat(RedisModuleIO *io, float value);
In the context of the rdb_save method of a module data type, saves a float
value to the RDB file. The float can be a valid number, a NaN or infinity.
It is possible to load back the value with `RedisModule_LoadFloat()`.
## `RM_LoadFloat`
float RM_LoadFloat(RedisModuleIO *io);
In the context of the rdb_save method of a module data type, loads back the
float value saved by `RedisModule_SaveFloat()`.
## `RM_EmitAOF`
void RM_EmitAOF(RedisModuleIO *io, const char *cmdname, const char *fmt, ...);
Emits a command into the AOF during the AOF rewriting process. This function
is only called in the context of the aof_rewrite method of data types exported
by a module. The command works exactly like `RedisModule_Call()` in the way
the parameters are passed, but it does not return anything as the error
handling is performed by Redis itself.
## `RM_LogRaw`
void RM_LogRaw(RedisModule *module, const char *levelstr, const char *fmt, va_list ap);
This is the low level function implementing both:
`RM_Log()`
`RM_LogIOError()`
## `RM_Log`
void RM_Log(RedisModuleCtx *ctx, const char *levelstr, const char *fmt, ...);
/*
Produces a log message to the standard Redis log, the format accepts
printf-alike specifiers, while level is a string describing the log
level to use when emitting the log, and must be one of the following:
* "debug"
* "verbose"
* "notice"
* "warning"
If the specified log level is invalid, verbose is used by default.
There is a fixed limit to the length of the log line this function is able
to emit, this limti is not specified but is guaranteed to be more than
a few lines of text.
## `RM_LogIOError`
void RM_LogIOError(RedisModuleIO *io, const char *levelstr, const char *fmt, ...);
Log errors from RDB / AOF serialization callbacks.
This function should be used when a callback is returning a critical
error to the caller since cannot load or save the data for some
critical reason.
## `RM_BlockClient`
RedisModuleBlockedClient *RM_BlockClient(RedisModuleCtx *ctx, RedisModuleCmdFunc reply_callback, RedisModuleCmdFunc timeout_callback, void (*free_privdata)(void*), long long timeout_ms);
Block a client in the context of a blocking command, returning an handle
which will be used, later, in order to block the client with a call to
`RedisModule_UnblockClient()`. The arguments specify callback functions
and a timeout after which the client is unblocked.
The callbacks are called in the following contexts:
reply_callback: called after a successful `RedisModule_UnblockClient()` call
in order to reply to the client and unblock it.
reply_timeout: called when the timeout is reached in order to send an
error to the client.
free_privdata: called in order to free the privata data that is passed
by `RedisModule_UnblockClient()` call.
## `RM_UnblockClient`
int RM_UnblockClient(RedisModuleBlockedClient *bc, void *privdata);
Unblock a client blocked by ``RedisModule_BlockedClient``. This will trigger
the reply callbacks to be called in order to reply to the client.
The 'privdata' argument will be accessible by the reply callback, so
the caller of this function can pass any value that is needed in order to
actually reply to the client.
A common usage for 'privdata' is a thread that computes something that
needs to be passed to the client, included but not limited some slow
to compute reply or some reply obtained via networking.
Note: this function can be called from threads spawned by the module.
## `RM_AbortBlock`
int RM_AbortBlock(RedisModuleBlockedClient *bc);
Abort a blocked client blocking operation: the client will be unblocked
without firing the reply callback.
## `RM_IsBlockedReplyRequest`
int RM_IsBlockedReplyRequest(RedisModuleCtx *ctx);
Return non-zero if a module command was called in order to fill the
reply for a blocked client.
## `RM_IsBlockedTimeoutRequest`
int RM_IsBlockedTimeoutRequest(RedisModuleCtx *ctx);
Return non-zero if a module command was called in order to fill the
reply for a blocked client that timed out.
## `RM_GetBlockedClientPrivateData`
void *RM_GetBlockedClientPrivateData(RedisModuleCtx *ctx);
Get the privata data set by `RedisModule_UnblockClient()`
Blocking commands in Redis modules
===
Redis has a few blocking commands among the built-in set of commands.
One of the most used is `BLPOP` (or the symmetric `BRPOP`) which blocks
waiting for elements arriving in a list.
The interesting fact about blocking commands is that they do not block
the whole server, but just the client calling them. Usually the reason to
block is that we expect some external event to happen: this can be
some change in the Redis data structures like in the `BLPOP` case, a
long computation happening in a thread, to receive some data from the
network, and so forth.
Redis modules have the ability to implement blocking commands as well,
this documentation shows how the API works and describes a few patterns
that can be used in order to model blocking commands.
How blocking and resuming works.
---
_Note: You may want to check the `helloblock.c` example in the Redis source tree
inside the `src/modules` directory, for a simple to understand example
on how the blocking API is applied._
In Redis modules, commands are implemented by callback functions that
are invoked by the Redis core when the specific command is called
by the user. Normally the callback terminates its execution sending
some reply to the client. Using the following function instead, the
function implementing the module command may request that the client
is put into the blocked state:
RedisModuleBlockedClient *RedisModule_BlockClient(RedisModuleCtx *ctx, RedisModuleCmdFunc reply_callback, RedisModuleCmdFunc timeout_callback, void (*free_privdata)(void*), long long timeout_ms);
The function returns a `RedisModuleBlockedClient` object, which is later
used in order to unblock the client. The arguments have the following
meaning:
* `ctx` is the command execution context as usually in the rest of the API.
* `reply_callback` is the callback, having the same prototype of a normal command function, that is called when the client is unblocked in order to return a reply to the client.
* `timeout_callback` is the callback, having the same prototype of a normal command function that is called when the client reached the `ms` timeout.
* `free_privdata` is the callback that is called in order to free the private data. Private data is a pointer to some data that is passed between the API used to unblock the client, to the callback that will send the reply to the client. We'll see how this mechanism works later in this document.
* `ms` is the timeout in milliseconds. When the timeout is reached, the timeout callback is called and the client is automatically aborted.
Once a client is blocked, it can be unblocked with the following API:
int RedisModule_UnblockClient(RedisModuleBlockedClient *bc, void *privdata);
The function takes as argument the blocked client object returned by
the previous call to `RedisModule_BlockClient()`, and unblock the client.
Immediately before the client gets unblocked, the `reply_callback` function
specified when the client was blocked is called: this function will
have access to the `privdata` pointer used here.
IMPORTANT: The above function is thread safe, and can be called from within
a thread doing some work in order to implement the command that blocked
the client.
The `privdata` data will be freed automatically using the `free_privdata`
callback when the client is unblocked. This is useful **since the reply
callback may never be called** in case the client timeouts or disconnects
from the server, so it's important that it's up to an external function
to have the responsibility to free the data passed if needed.
To better understand how the API works, we can imagine writing a command
that blocks a client for one second, and then send as reply "Hello!".
Note: arity checks and other non important things are not implemented
int his command, in order to take the example simple.
int Example_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv,
int argc)
{
RedisModuleBlockedClient *bc =
RedisModule_BlockClient(ctx,reply_func,timeout_func,NULL,0);
pthread_t tid;
pthread_create(&tid,NULL,threadmain,bc);
return REDISMODULE_OK;
}
void *threadmain(void *arg) {
RedisModuleBlockedClient *bc = arg;
sleep(1); /* Wait one second and unblock. */
RedisModule_UnblockClient(bc,NULL);
}
The above command blocks the client ASAP, spawining a thread that will
wait a second and will unblock the client. Let's check the reply and
timeout callbacks, which are in our case very similar, since they
just reply the client with a different reply type.
int reply_func(RedisModuleCtx *ctx, RedisModuleString **argv,
int argc)
{
return RedisModule_ReplyWithSimpleString(ctx,"Hello!");
}
int timeout_func(RedisModuleCtx *ctx, RedisModuleString **argv,
int argc)
{
return RedisModule_ReplyWithNull(ctx);
}
The reply callback just sends the "Hello!" string to the client.
The important bit here is that the reply callback is called when the
client is unblocked from the thread.
The timeout command returns `NULL`, as it often happens with actual
Redis blocking commands timing out.
Passing reply data when unblocking
---
The above example is simple to understand but lacks an important
real world aspect of an actual blocking command implementation: often
the reply function will need to know what to reply to the client,
and this information is often provided as the client is unblocked.
We could modify the above example so that the thread generates a
random number after waiting one second. You can think at it as an
actually expansive operation of some kind. Then this random number
can be passed to the reply function so that we return it to the command
caller. In order to make this working, we modify the functions as follow:
void *threadmain(void *arg) {
RedisModuleBlockedClient *bc = arg;
sleep(1); /* Wait one second and unblock. */
long *mynumber = RedisModule_Alloc(sizeof(long));
*mynumber = rand();
RedisModule_UnblockClient(bc,mynumber);
}
As you can see, now the unblocking call is passing some private data,
that is the `mynumber` pointer, to the reply callback. In order to
obtain this private data, the reply callback will use the following
fnuction:
void *RedisModule_GetBlockedClientPrivateData(RedisModuleCtx *ctx);
So our reply callback is modified like that:
int reply_func(RedisModuleCtx *ctx, RedisModuleString **argv,
int argc)
{
long *mynumber = RedisModule_GetBlockedClientPrivateData(ctx);
/* IMPORTANT: don't free mynumber here, but in the
* free privdata callback. */
return RedisModule_ReplyWithLongLong(ctx,mynumber);
}
Note that we also need to pass a `free_privdata` function when blocking
the client with `RedisModule_BlockClient()`, since the allocated
long value must be freed. Our callback will look like the following:
void free_privdata(void *privdata) {
RedisModule_Free(privdata);
}
NOTE: It is important to stress that the private data is best freed in the
`free_privdata` callback becaues the reply function may not be called
if the client disconnects or timeout.
Also note that the private data is also accessible from the timeout
callback, always using the `GetBlockedClientPrivateData()` API.
Aborting the blocking of a client
---
One problem that sometimes arises is that we need to allocate resources
in order to implement the non blocking command. So we block the client,
then, for example, try to create a thread, but the thread creation function
returns an error. What to do in such a condition in order to recover? We
don't want to take the client blocked, nor we want to call `UnblockClient()`
because this will trigger the reply callback to be called.
In this case the best thing to do is to use the following function:
int RedisModule_AbortBlock(RedisModuleBlockedClient *bc);
Practically this is how to use it:
int Example_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv,
int argc)
{
RedisModuleBlockedClient *bc =
RedisModule_BlockClient(ctx,reply_func,timeout_func,NULL,0);
pthread_t tid;
if (pthread_create(&tid,NULL,threadmain,bc) != 0) {
RedisModule_AbortBlock(bc);
RedisModule_ReplyWithError(ctx,"Sorry can't create a thread");
}
return REDISMODULE_OK;
}
The client will be unblocked but the reply callback will not be called.
Implementing the command, reply and timeout callback using a single function
---
The following functions can be used in order to implement the reply and
callback with the same function that implements the primary command
function:
int RedisModule_IsBlockedReplyRequest(RedisModuleCtx *ctx);
int RedisModule_IsBlockedTimeoutRequest(RedisModuleCtx *ctx);
So I could rewrite the example command without using a separated
reply and timeout callback:
int Example_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv,
int argc)
{
if (RedisModule_IsBlockedReplyRequest(ctx)) {
long *mynumber = RedisModule_GetBlockedClientPrivateData(ctx);
return RedisModule_ReplyWithLongLong(ctx,mynumber);
} else if (RedisModule_IsBlockedTimeoutRequest) {
return RedisModule_ReplyWithNull(ctx);
}
RedisModuleBlockedClient *bc =
RedisModule_BlockClient(ctx,reply_func,timeout_func,NULL,0);
pthread_t tid;
if (pthread_create(&tid,NULL,threadmain,bc) != 0) {
RedisModule_AbortBlock(bc);
RedisModule_ReplyWithError(ctx,"Sorry can't create a thread");
}
return REDISMODULE_OK;
}
Functionally is the same but there are people that will prefer the less
verbose implementation that concentrates most of the command logic in a
single function.
Working on copies of data inside a thread
---
An interesting pattern in order to work with threads implementing the
slow part of a command, is to work with a copy of the data, so that
while some operation is performed in a key, the user continues to see
the old version. However when the thread terminated its work, the
representations are swapped and the new, processed version, is used.
An example of this approach is the
[Neural Redis module](https://github.com/antirez/neural-redis)
where neural networks are trained in different threads while the
user can still execute and inspect their older versions.
Future work
---
An API is work in progress right now in order to allow Redis modules APIs
to be called in a safe way from threads, so that the threaded command
can access the data space and do incremental operations.
There is no ETA for this feature but it may appear in the course of the
Redis 4.0 release at some point.
Redis Modules: an introduction to the API
===
The modules documentation is composed of the following files:
* `INTRO.md` (this file). An overview about Redis Modules system and API. It's a good idea to start your reading here.
* `API.md` is generated from module.c top comments of RedisMoule functions. It is a good reference in order to understand how each function works.
* `TYPES.md` covers the implementation of native data types into modules.
* `BLOCK.md` shows how to write blocking commands that will not reply immediately, but will block the client, without blocking the Redis server, and will provide a reply whenever will be possible.
Redis modules make possible to extend Redis functionality using external
modules, implementing new Redis commands at a speed and with features
similar to what can be done inside the core itself.
Redis modules are dynamic libraries, that can be loaded into Redis at
startup or using the `MODULE LOAD` command. Redis exports a C API, in the
form of a single C header file called `redismodule.h`. Modules are meant
to be written in C, however it will be possible to use C++ or other languages
that have C binding functionalities.
Modules are designed in order to be loaded into different versions of Redis,
so a given module does not need to be designed, or recompiled, in order to
run with a specific version of Redis. For this reason, the module will
register to the Redis core using a specific API version. The current API
version is "1".
This document is about an alpha version of Redis modules. API, functionalities
and other details may change in the future.
# Loading modules
In order to test the module you are developing, you can load the module
using the following `redis.conf` configuration directive:
loadmodule /path/to/mymodule.so
It is also possible to load a module at runtime using the following command:
MODULE LOAD /path/to/mymodule.so
In order to list all loaded modules, use:
MODULE LIST
Finally, you can unload (and later reload if you wish) a module using the
following command:
MODULE UNLOAD mymodule
Note that `mymodule` above is not the filename without the `.so` suffix, but
instead, the name the module used to register itself into the Redis core.
The name can be obtained using `MODULE LIST`. However it is good practice
that the filename of the dynamic library is the same as the name the module
uses to register itself into the Redis core.
# The simplest module you can write
In order to show the different parts of a module, here we'll show a very
simple module that implements a command that outputs a random number.
#include "redismodule.h"
#include <stdlib.h>
int HelloworldRand_RedisCommand(RedisModuleCtx *ctx, RedisModuleString **argv, int argc) {
RedisModule_ReplyWithLongLong(ctx,rand());
return REDISMODULE_OK;
}
int RedisModule_OnLoad(RedisModuleCtx *ctx, RedisModuleString **argv, int argc) {
if (RedisModule_Init(ctx,"helloworld",1,REDISMODULE_APIVER_1)
== REDISMODULE_ERR) return REDISMODULE_ERR;
if (RedisModule_CreateCommand(ctx,"helloworld.rand",
HelloworldRand_RedisCommand) == REDISMODULE_ERR)
return REDISMODULE_ERR;
return REDISMODULE_OK;
}
The example module has two functions. One implements a command called
HELLOWORLD.RAND. This function is specific of that module. However the
other function called `RedisModule_OnLoad()` must be present in each
Redis module. It is the entry point for the module to be initialized,
register its commands, and potentially other private data structures
it uses.
Note that it is a good idea for modules to call commands with the
name of the module followed by a dot, and finally the command name,
like in the case of `HELLOWORLD.RAND`. This way it is less likely to
have collisions.
Note that if different modules have colliding commands, they'll not be
able to work in Redis at the same time, since the function
`RedisModule_CreateCommand` will fail in one of the modules, so the module
loading will abort returning an error condition.
# Module initialization
The above example shows the usage of the function `RedisModule_Init()`.
It should be the first function called by the module `OnLoad` function.
The following is the function prototype:
int RedisModule_Init(RedisModuleCtx *ctx, const char *modulename,
int module_version, int api_version);
The `Init` function announces the Redis core that the module has a given
name, its version (that is reported by `MODULE LIST`), and that is willing
to use a specific version of the API.
If the API version is wrong, the name is already taken, or there are other
similar errors, the function will return `REDISMODULE_ERR`, and the module
`OnLoad` function should return ASAP with an error.
Before the `Init` function is called, no other API function can be called,
otherwise the module will segfault and the Redis instance will crash.
The second function called, `RedisModule_CreateCommand`, is used in order
to register commands into the Redis core. The following is the prototype:
int RedisModule_CreateCommand(RedisModuleCtx *ctx, const char *cmdname,
RedisModuleCmdFunc cmdfunc);
As you can see, most Redis modules API calls all take as first argument
the `context` of the module, so that they have a reference to the module
calling it, to the command and client executing a given command, and so forth.
To create a new command, the above function needs the context, the command
name, and the function pointer of the function implementing the command,
which must have the following prototype:
int mycommand(RedisModuleCtx *ctx, RedisModuleString **argv, int argc);
The command function arguments are just the context, that will be passed
to all the other API calls, the command argument vector, and total number
of arguments, as passed by the user.
As you can see, the arguments are provided as pointers to a specific data
type, the `RedisModuleString`. This is an opaque data type you have API
functions to access and use, direct access to its fields is never needed.
Zooming into the example command implementation, we can find another call:
int RedisModule_ReplyWithLongLong(RedisModuleCtx *ctx, long long integer);
This function returns an integer to the client that invoked the command,
exactly like other Redis commands do, like for example `INCR` or `SCARD`.
# Setup and dependencies of a Redis module
Redis modules don't depend on Redis or some other library, nor they
need to be compiled with a specific `redismodule.h` file. In order
to create a new module, just copy a recent version of `redismodule.h`
in your source tree, link all the libraries you want, and create
a dynamic library having the `RedisModule_OnLoad()` function symbol
exported.
The module will be able to load into different versions of Redis.
# Passing configuration parameters to Redis modules
When the module is loaded with the `MODULE LOAD` command, or using the
`loadmodule` directive in the `redis.conf` file, the user is able to pass
configuration parameters to the module by adding arguments after the module
file name:
loadmodule mymodule.so foo bar 1234
In the above example the strings `foo`, `bar` and `123` will be passed
to the module `OnLoad()` function in the `argv` argument as an array
of RedisModuleString pointers. The number of arguments passed is into `argc`.
The way you can access those strings will be explained in the rest of this
document. Normally the module will store the module configuration parameters
in some `static` global variable that can be accessed module wide, so that
the configuration can change the behavior of different commands.
# Working with RedisModuleString objects
The command argument vector `argv` passed to module commands, and the
return value of other module APIs functions, are of type `RedisModuleString`.
Usually you directly pass module strings to other API calls, however sometimes
you may need to directly access the string object.
There are a few functions in order to work with string objects:
const char *RedisModule_StringPtrLen(RedisModuleString *string, size_t *len);
The above function accesses a string by returning its pointer and setting its
length in `len`.
You should never write to a string object pointer, as you can see from the
`const` pointer qualifier.
However, if you want, you can create new string objects using the following
API:
RedisModuleString *RedisModule_CreateString(RedisModuleCtx *ctx, const char *ptr, size_t len);
The string returned by the above command must be freed using a corresponding
call to `RedisModule_FreeString()`:
void RedisModule_FreeString(RedisModuleString *str);
However if you want to avoid having to free strings, the automatic memory
management, covered later in this document, can be a good alternative, by
doing it for you.
Note that the strings provided via the argument vector `argv` never need
to be freed. You only need to free new strings you create, or new strings
returned by other APIs, where it is specified that the returned string must
be freed.
## Creating strings from numbers or parsing strings as numbers
Creating a new string from an integer is a very common operation, so there
is a function to do this:
RedisModuleString *mystr = RedisModule_CreateStringFromLongLong(ctx,10);
Similarly in order to parse a string as a number:
long long myval;
if (RedisModule_StringToLongLong(ctx,argv[1],&myval) == REDISMODULE_OK) {
/* Do something with 'myval' */
}
## Accessing Redis keys from modules
Most Redis modules, in order to be useful, have to interact with the Redis
data space (this is not always true, for example an ID generator may
never touch Redis keys). Redis modules have two different APIs in order to
access the Redis data space, one is a low level API that provides very
fast access and a set of functions to manipulate Redis data structures.
The other API is more high level, and allows to call Redis commands and
fetch the result, similarly to how Lua scripts access Redis.
The high level API is also useful in order to access Redis functionalities
that are not available as APIs.
In general modules developers should prefer the low level API, because commands
implemented using the low level API run at a speed comparable to the speed
of native Redis commands. However there are definitely use cases for the
higher level API. For example often the bottleneck could be processing the
data and not accessing it.
Also note that sometimes using the low level API is not harder compared to
the higher level one.
# Calling Redis commands
The high level API to access Redis is the sum of the `RedisModule_Call()`
function, together with the functions needed in order to access the
reply object returned by `Call()`.
`RedisModule_Call` uses a special calling convention, with a format specifier
that is used to specify what kind of objects you are passing as arguments
to the function.
Redis commands are invoked just using a command name and a list of arguments.
However when calling commands, the arguments may originate from different
kind of strings: null-terminated C strings, RedisModuleString objects as
received from the `argv` parameter in the command implementation, binary
safe C buffers with a pointer and a length, and so forth.
For example if I want to call `INCRBY` using a first argument (the key)
a string received in the argument vector `argv`, which is an array
of RedisModuleString object pointers, and a C string representing the
number "10" as second argument (the increment), I'll use the following
function call:
RedisModuleCallReply *reply;
reply = RedisModule_Call(ctx,"INCR","sc",argv[1],"10");
The first argument is the context, and the second is always a null terminated
C string with the command name. The third argument is the format specifier
where each character corresponds to the type of the arguments that will follow.
In the above case `"sc"` means a RedisModuleString object, and a null
terminated C string. The other arguments are just the two arguments as
specified. In fact `argv[1]` is a RedisModuleString and `"10"` is a null
terminated C string.
This is the full list of format specifiers:
* **c** -- Null terminated C string pointer.
* **b** -- C buffer, two arguments needed: C string pointer and `size_t` length.
* **s** -- RedisModuleString as received in `argv` or by other Redis module APIs returning a RedisModuleString object.
* **l** -- Long long integer.
* **v** -- Array of RedisModuleString objects.
* **!** -- This modifier just tells the function to replicate the command to slaves and AOF. It is ignored from the point of view of arguments parsing.
The function returns a `RedisModuleCallReply` object on success, on
error NULL is returned.
NULL is returned when the command name is invalid, the format specifier uses
characters that are not recognized, or when the command is called with the
wrong number of arguments. In the above cases the `errno` var is set to `EINVAL`. NULL is also returned when, in an instance with Cluster enabled, the target
keys are about non local hash slots. In this case `errno` is set to `EPERM`.
## Working with RedisModuleCallReply objects.
`RedisModuleCall` returns reply objects that can be accessed using the
`RedisModule_CallReply*` family of functions.
In order to obtain the type or reply (corresponding to one of the data types
supported by the Redis protocol), the function `RedisModule_CallReplyType()`
is used:
reply = RedisModule_Call(ctx,"INCR","sc",argv[1],"10");
if (RedisModule_CallReplyType(reply) == REDISMODULE_REPLY_INTEGER) {
long long myval = RedisModule_CallReplyInteger(reply);
/* Do something with myval. */
}
Valid reply types are:
* `REDISMODULE_REPLY_STRING` Bulk string or status replies.
* `REDISMODULE_REPLY_ERROR` Errors.
* `REDISMODULE_REPLY_INTEGER` Signed 64 bit integers.
* `REDISMODULE_REPLY_ARRAY` Array of replies.
* `REDISMODULE_REPLY_NULL` NULL reply.
Strings, errors and arrays have an associated length. For strings and errors
the length corresponds to the length of the string. For arrays the length
is the number of elements. To obtain the reply length the following function
is used:
size_t reply_len = RedisModule_CallReplyLength(reply);
In order to obtain the value of an integer reply, the following function is used, as already shown in the example above:
long long reply_integer_val = RedisModule_CallReplyInteger(reply);
Called with a reply object of the wrong type, the above function always
returns `LLONG_MIN`.
Sub elements of array replies are accessed this way:
RedisModuleCallReply *subreply;
subreply = RedisModule_CallReplyArrayElement(reply,idx);
The above function returns NULL if you try to access out of range elements.
Strings and errors (which are like strings but with a different type) can
be accessed using in the following way, making sure to never write to
the resulting pointer (that is returned as as `const` pointer so that
misusing must be pretty explicit):
size_t len;
char *ptr = RedisModule_CallReplyStringPtr(reply,&len);
If the reply type is not a string or an error, NULL is returned.
RedisCallReply objects are not the same as module string objects
(RedisModuleString types). However sometimes you may need to pass replies
of type string or integer, to API functions expecting a module string.
When this is the case, you may want to evaluate if using the low level
API could be a simpler way to implement your command, or you can use
the following function in order to create a new string object from a
call reply of type string, error or integer:
RedisModuleString *mystr = RedisModule_CreateStringFromCallReply(myreply);
If the reply is not of the right type, NULL is returned.
The returned string object should be released with `RedisModule_FreeString()`
as usually, or by enabling automatic memory management (see corresponding
section).
# Releasing call reply objects
Reply objects must be freed using `RedisModule_FreeCallReply`. For arrays,
you need to free only the top level reply, not the nested replies.
Currently the module implementation provides a protection in order to avoid
crashing if you free a nested reply object for error, however this feature
is not guaranteed to be here forever, so should not be considered part
of the API.
If you use automatic memory management (explained later in this document)
you don't need to free replies (but you still could if you wish to release
memory ASAP).
## Returning values from Redis commands
Like normal Redis commands, new commands implemented via modules must be
able to return values to the caller. The API exports a set of functions for
this goal, in order to return the usual types of the Redis protocol, and
arrays of such types as elemented. Also errors can be returned with any
error string and code (the error code is the initial uppercase letters in
the error message, like the "BUSY" string in the "BUSY the sever is busy" error
message).
All the functions to send a reply to the client are called
`RedisModule_ReplyWith<something>`.
To return an error, use:
RedisModule_ReplyWithError(RedisModuleCtx *ctx, const char *err);
There is a predefined error string for key of wrong type errors:
REDISMODULE_ERRORMSG_WRONGTYPE
Example usage:
RedisModule_ReplyWithError(ctx,"ERR invalid arguments");
We already saw how to reply with a long long in the examples above:
RedisModule_ReplyWithLongLong(ctx,12345);
To reply with a simple string, that can't contain binary values or newlines,
(so it's suitable to send small words, like "OK") we use:
RedisModule_ReplyWithSimpleString(ctx,"OK");
It's possible to reply with "bulk strings" that are binary safe, using
two different functions:
int RedisModule_ReplyWithStringBuffer(RedisModuleCtx *ctx, const char *buf, size_t len);
int RedisModule_ReplyWithString(RedisModuleCtx *ctx, RedisModuleString *str);
The first function gets a C pointer and length. The second a RedisMoudleString
object. Use one or the other depending on the source type you have at hand.
In order to reply with an array, you just need to use a function to emit the
array length, followed by as many calls to the above functions as the number
of elements of the array are:
RedisModule_ReplyWithArray(ctx,2);
RedisModule_ReplyWithStringBuffer(ctx,"age",3);
RedisModule_ReplyWithLongLong(ctx,22);
To return nested arrays is easy, your nested array element just uses another
call to `RedisModule_ReplyWithArray()` followed by the calls to emit the
sub array elements.
## Returning arrays with dynamic length
Sometimes it is not possible to know beforehand the number of items of
an array. As an example, think of a Redis module implementing a FACTOR
command that given a number outputs the prime factors. Instead of
factorializing the number, storing the prime factors into an array, and
later produce the command reply, a better solution is to start an array
reply where the length is not known, and set it later. This is accomplished
with a special argument to `RedisModule_ReplyWithArray()`:
RedisModule_ReplyWithArray(ctx, REDISMODULE_POSTPONED_ARRAY_LEN);
The above call starts an array reply so we can use other `ReplyWith` calls
in order to produce the array items. Finally in order to set the length
se use the following call:
RedisModule_ReplySetArrayLength(ctx, number_of_items);
In the case of the FACTOR command, this translates to some code similar
to this:
RedisModule_ReplyWithArray(ctx, REDISMODULE_POSTPONED_ARRAY_LEN);
number_of_factors = 0;
while(still_factors) {
RedisModule_ReplyWithLongLong(ctx, some_factor);
number_of_factors++;
}
RedisModule_ReplySetArrayLength(ctx, number_of_factors);
Another common use case for this feature is iterating over the arrays of
some collection and only returning the ones passing some kind of filtering.
It is possible to have multiple nested arrays with postponed reply.
Each call to `SetArray()` will set the length of the latest corresponding
call to `ReplyWithArray()`:
RedisModule_ReplyWithArray(ctx, REDISMODULE_POSTPONED_ARRAY_LEN);
... generate 100 elements ...
RedisModule_ReplyWithArray(ctx, REDISMODULE_POSTPONED_ARRAY_LEN);
... generate 10 elements ...
RedisModule_ReplySetArrayLength(ctx, 10);
RedisModule_ReplySetArrayLength(ctx, 100);
This creates a 100 items array having as last element a 10 items array.
# Arity and type checks
Often commands need to check that the number of arguments and type of the key
is correct. In order to report a wrong arity, there is a specific function
called `RedisModule_WrongArity()`. The usage is trivial:
if (argc != 2) return RedisModule_WrongArity(ctx);
Checking for the wrong type involves opening the key and checking the type:
RedisModuleKey *key = RedisModule_OpenKey(ctx,argv[1],
REDISMODULE_READ|REDISMODULE_WRITE);
int keytype = RedisModule_KeyType(key);
if (keytype != REDISMODULE_KEYTYPE_STRING &&
keytype != REDISMODULE_KEYTYPE_EMPTY)
{
RedisModule_CloseKey(key);
return RedisModule_ReplyWithError(ctx,REDISMODULE_ERRORMSG_WRONGTYPE);
}
Note that you often want to proceed with a command both if the key
is of the expected type, or if it's empty.
## Low level access to keys
Low level access to keys allow to perform operations on value objects associated
to keys directly, with a speed similar to what Redis uses internally to
implement the built-in commands.
Once a key is opened, a key pointer is returned that will be used with all the
other low level API calls in order to perform operations on the key or its
associated value.
Because the API is meant to be very fast, it cannot do too many run-time
checks, so the user must be aware of certain rules to follow:
* Opening the same key multiple times where at least one instance is opened for writing, is undefined and may lead to crashes.
* While a key is open, it should only be accessed via the low level key API. For example opening a key, then calling DEL on the same key using the `RedisModule_Call()` API will result into a crash. However it is safe to open a key, perform some operation with the low level API, closing it, then using other APIs to manage the same key, and later opening it again to do some more work.
In order to open a key the `RedisModule_OpenKey` function is used. It returns
a key pointer, that we'll use with all the next calls to access and modify
the value:
RedisModuleKey *key;
key = RedisModule_OpenKey(ctx,argv[1],REDISMODULE_READ);
The second argument is the key name, that must be a `RedisModuleString` object.
The third argument is the mode: `REDISMODULE_READ` or `REDISMODULE_WRITE`.
It is possible to use `|` to bitwise OR the two modes to open the key in
both modes. Currently a key opened for writing can also be accessed for reading
but this is to be considered an implementation detail. The right mode should
be used in sane modules.
You can open non exisitng keys for writing, since the keys will be created
when an attempt to write to the key is performed. However when opening keys
just for reading, `RedisModule_OpenKey` will return NULL if the key does not
exist.
Once you are done using a key, you can close it with:
RedisModule_CloseKey(key);
Note that if automatic memory management is enabled, you are not forced to
close keys. When the module function returns, Redis will take care to close
all the keys which are still open.
## Getting the key type
In order to obtain the value of a key, use the `RedisModule_KeyType()` function:
int keytype = RedisModule_KeyType(key);
It returns one of the following values:
REDISMODULE_KEYTYPE_EMPTY
REDISMODULE_KEYTYPE_STRING
REDISMODULE_KEYTYPE_LIST
REDISMODULE_KEYTYPE_HASH
REDISMODULE_KEYTYPE_SET
REDISMODULE_KEYTYPE_ZSET
The above are just the usual Redis key types, with the addition of an empty
type, that signals the key pointer is associated with an empty key that
does not yet exists.
## Creating new keys
To create a new key, open it for writing and then write to it using one
of the key writing functions. Example:
RedisModuleKey *key;
key = RedisModule_OpenKey(ctx,argv[1],REDISMODULE_READ);
if (RedisModule_KeyType(key) == REDISMODULE_KEYTYPE_EMPTY) {
RedisModule_StringSet(key,argv[2]);
}
## Deleting keys
Just use:
RedisModule_DeleteKey(key);
The function returns `REDISMODULE_ERR` if the key is not open for writing.
Note that after a key gets deleted, it is setup in order to be targeted
by new key commands. For example `RedisModule_KeyType()` will return it is
an empty key, and writing to it will create a new key, possibly of another
type (depending on the API used).
## Managing key expires (TTLs)
To control key expires two functions are provided, that are able to set,
modify, get, and unset the time to live associated with a key.
One function is used in order to query the current expire of an open key:
mstime_t RedisModule_GetExpire(RedisModuleKey *key);
The function returns the time to live of the key in milliseconds, or
`REDISMODULE_NO_EXPIRE` as a special value to signal the key has no associated
expire or does not exist at all (you can differentiate the two cases checking
if the key type is `REDISMODULE_KEYTYPE_EMPTY`).
In order to change the expire of a key the following function is used instead:
int RedisModule_SetExpire(RedisModuleKey *key, mstime_t expire);
When called on a non existing key, `REDISMODULE_ERR` is returned, because
the function can only associate expires to existing open keys (non existing
open keys are only useful in order to create new values with data type
specific write operations).
Again the `expire` time is specified in milliseconds. If the key has currently
no expire, a new expire is set. If the key already have an expire, it is
replaced with the new value.
If the key has an expire, and the special value `REDISMODULE_NO_EXPIRE` is
used as a new expire, the expire is removed, similarly to the Redis
`PERSIST` command. In case the key was already persistent, no operation is
performed.
## Obtaining the length of values
There is a single function in order to retrieve the length of the value
associated to an open key. The returned length is value-specific, and is
the string length for strings, and the number of elements for the aggregated
data types (how many elements there is in a list, set, sorted set, hash).
size_t len = RedisModule_ValueLength(key);
If the key does not exist, 0 is returned by the function:
## String type API
Setting a new string value, like the Redis `SET` command does, is performed
using:
int RedisModule_StringSet(RedisModuleKey *key, RedisModuleString *str);
The function works exactly like the Redis `SET` command itself, that is, if
there is a prior value (of any type) it will be deleted.
Accessing existing string values is performed using DMA (direct memory
access) for speed. The API will return a pointer and a length, so that's
possible to access and, if needed, modify the string directly.
size_t len, j;
char *myptr = RedisModule_StringDMA(key,&len,REDISMODULE_WRITE);
for (j = 0; j < len; j++) myptr[j] = 'A';
In the above example we write directly on the string. Note that if you want
to write, you must be sure to ask for `WRITE` mode.
DMA pointers are only valid if no other operations are performed with the key
before using the pointer, after the DMA call.
Sometimes when we want to manipulate strings directly, we need to change
their size as well. For this scope, the `RedisModule_StringTruncate` function
is used. Example:
RedisModule_StringTruncate(mykey,1024);
The function truncates, or enlarges the string as needed, padding it with
zero bytes if the previos length is smaller than the new length we request.
If the string does not exist since `key` is associated to an open empty key,
a string value is created and associated to the key.
Note that every time `StringTruncate()` is called, we need to re-obtain
the DMA pointer again, since the old may be invalid.
## List type API
It's possible to push and pop values from list values:
int RedisModule_ListPush(RedisModuleKey *key, int where, RedisModuleString *ele);
RedisModuleString *RedisModule_ListPop(RedisModuleKey *key, int where);
In both the APIs the `where` argument specifies if to push or pop from tail
or head, using the following macros:
REDISMODULE_LIST_HEAD
REDISMODULE_LIST_TAIL
Elements returned by `RedisModule_ListPop()` are like strings craeted with
`RedisModule_CreateString()`, they must be released with
`RedisModule_FreeString()` or by enabling automatic memory management.
## Set type API
Work in progress.
## Sorted set type API
Documentation missing, please refer to the top comments inside `module.c`
for the following functions:
* `RedisModule_ZsetAdd`
* `RedisModule_ZsetIncrby`
* `RedisModule_ZsetScore`
* `RedisModule_ZsetRem`
And for the sorted set iterator:
* `RedisModule_ZsetRangeStop`
* `RedisModule_ZsetFirstInScoreRange`
* `RedisModule_ZsetLastInScoreRange`
* `RedisModule_ZsetFirstInLexRange`
* `RedisModule_ZsetLastInLexRange`
* `RedisModule_ZsetRangeCurrentElement`
* `RedisModule_ZsetRangeNext`
* `RedisModule_ZsetRangePrev`
* `RedisModule_ZsetRangeEndReached`
## Hash type API
Documentation missing, please refer to the top comments inside `module.c`
for the following functions:
* `RedisModule_HashSet`
* `RedisModule_HashGet`
## Iterating aggregated values
Work in progress.
# Replicating commands
If you want to use module commands exactly like normal Redis commands, in the
context of replicated Redis instances, or using the AOF file for persistence,
it is important for module commands to handle their replication in a consistent
way.
When using the higher level APIs to invoke commands, replication happens
automatically if you use the "!" modifier in the format string of
`RedisModule_Call()` as in the following example:
reply = RedisModule_Call(ctx,"INCR","!sc",argv[1],"10");
As you can see the format specifier is `"!sc"`. The bang is not parsed as a
format specifier, but it internally flags the command as "must replicate".
If you use the above programming style, there are no problems.
However sometimes things are more complex than that, and you use the low level
API. In this case, if there are no side effects in the command execution, and
it consistently always performs the same work, what is possible to do is to
replicate the command verbatim as the user executed it. To do that, you just
need to call the following function:
RedisModule_ReplicateVerbatim(ctx);
When you use the above API, you should not use any other replication function
since they are not guaranteed to mix well.
However this is not the only option. It's also possible to exactly tell
Redis what commands to replicate as the effect of the command execution, using
an API similar to `RedisModule_Call()` but that instead of calling the command
sends it to the AOF / slaves stream. Example:
RedisModule_Replicate(ctx,"INCRBY","cl","foo",my_increment);
It's possible to call `RedisModule_Replicate` multiple times, and each
will emit a command. All the sequence emitted is wrapped between a
`MULTI/EXEC` transaction, so that the AOF and replication effects are the
same as executing a single command.
Note that `Call()` replication and `Replicate()` replication have a rule,
in case you want to mix both forms of replication (not necessarily a good
idea if there are simpler approaches). Commands replicated with `Call()`
are always the first emitted in the final `MULTI/EXEC` block, while all
the commands emitted with `Replicate()` will follow.
# Automatic memory management
Normally when writing programs in the C language, programmers need to manage
memory manually. This is why the Redis modules API has functions to release
strings, close open keys, free replies, and so forth.
However given that commands are executed in a contained environment and
with a set of strict APIs, Redis is able to provide automatic memory management
to modules, at the cost of some performance (most of the time, a very low
cost).
When automatic memory management is enabled:
1. You don't need to close open keys.
2. You don't need to free replies.
3. You don't need to free RedisModuleString objects.
However you can still do it, if you want. For example, automatic memory
management may be active, but inside a loop allocating a lot of strings,
you may still want to free strings no longer used.
In order to enable automatic memory management, just call the following
function at the start of the command implementation:
RedisModule_AutoMemory(ctx);
Automatic memory management is usually the way to go, however experienced
C programmers may not use it in order to gain some speed and memory usage
benefit.
# Allocating memory into modules
Normal C programs use `malloc()` and `free()` in order to allocate and
release memory dynamically. While in Redis modules the use of malloc is
not technically forbidden, it is a lot better to use the Redis Modules
specific functions, that are exact replacements for `malloc`, `free`,
`realloc` and `strdup`. These functions are:
void *RedisModule_Alloc(size_t bytes);
void* RedisModule_Realloc(void *ptr, size_t bytes);
void RedisModule_Free(void *ptr);
void RedisModule_Calloc(size_t nmemb, size_t size);
char *RedisModule_Strdup(const char *str);
They work exactly like their `libc` equivalent calls, however they use
the same allocator Redis uses, and the memory allocated using these
functions is reported by the `INFO` command in the memory section, is
accounted when enforcing the `maxmemory` policy, and in general is
a first citizen of the Redis executable. On the contrar, the method
allocated inside modules with libc `malloc()` is transparent to Redis.
Another reason to use the modules functions in order to allocate memory
is that, when creating native data types inside modules, the RDB loading
functions can return deserialized strings (from the RDB file) directly
as `RedisModule_Alloc()` allocations, so they can be used directly to
populate data structures after loading, instead of having to copy them
to the data structure.
## Pool allocator
Sometimes in commands implementations, it is required to perform many
small allocations that will be not retained at the end of the command
execution, but are just functional to execute the command itself.
This work can be more easily accomplished using the Redis pool allocator:
void *RedisModule_PoolAlloc(RedisModuleCtx *ctx, size_t bytes);
It works similarly to `malloc()`, and returns memory aligned to the
next power of two of greater or equal to `bytes` (for a maximum alignment
of 8 bytes). However it allocates memory in blocks, so it the overhead
of the allocations is small, and more important, the memory allocated
is automatically released when the command returns.
So in general short living allocations are a good candidates for the pool
allocator.
# Writing commands compatible with Redis Cluster
Documentation missing, please check the following functions inside `module.c`:
RedisModule_IsKeysPositionRequest(ctx);
RedisModule_KeyAtPos(ctx,pos);
Native types in Redis modules
===
Redis modules can access Redis built-in data structures both at high level,
by calling Redis commands, and at low level, by manipulating the data structures
directly.
By using these capabilities in order to build new abstractions on top of existing
Redis data structures, or by using strings DMA in order to encode modules
data structures into Redis strings, it is possible to create modules that
*feel like* they are exporting new data types. However, for more complex
problems, this is not enough, and the implementation of new data structures
inside the module is needed.
We call the ability of Redis modules to implement new data structures that
feel like native Redis ones **native types support**. This document describes
the API exported by the Redis modules system in order to create new data
structures and handle the serialization in RDB files, the rewriting process
in AOF, the type reporting via the `TYPE` command, and so forth.
Overview of native types
---
A module exporting a native type is composed of the following main parts:
* The implementation of some kind of new data structure and of commands operating on the new data structure.
* A set of callbacks that handle: RDB saving, RDB loading, AOF rewriting, releasing of a value associated with a key, calculation of a value digest (hash) to be used with the `DEBUG DIGEST` command.
* A 9 characters name that is unique to each module native data type.
* An encoding version, used to persist into RDB files a module-specific data version, so that a module will be able to load older representations from RDB files.
While to handle RDB loading, saving and AOF rewriting may look complex as a first glance, the modules API provide very high level function for handling all this, without requiring the user to handle read/write errors, so in practical terms, writing a new data structure for Redis is a simple task.
A **very easy** to understand but complete example of native type implementation
is available inside the Redis distribution in the `/modules/hellotype.c` file.
The reader is encouraged to read the documentation by looking at this example
implementation to see how things are applied in the practice.
Registering a new data type
===
In order to register a new native type into the Redis core, the module needs
to declare a global variable that will hold a reference to the data type.
The API to register the data type will return a data type reference that will
be stored in the global variable.
static RedisModuleType *MyType;
#define MYTYPE_ENCODING_VERSION 0
int RedisModule_OnLoad(RedisModuleCtx *ctx) {
RedisModuleTypeMethods tm = {
.version = REDISMODULE_TYPE_METHOD_VERSION,
.rdb_load = MyTypeRDBLoad,
.rdb_save = MyTypeRDBSave,
.aof_rewrite = MyTypeAOFRewrite,
.free = MyTypeFree
};
MyType = RedisModule_CreateDataType(ctx, "MyType-AZ",
MYTYPE_ENCODING_VERSION, &tm);
if (MyType == NULL) return REDISMODULE_ERR;
}
As you can see from the example above, a single API call is needed in order to
register the new type. However a number of function pointers are passed as
arguments. Certain are optionals while some are mandatory. The above set
of methods *must* be passed, while `.digest` and `.mem_usage` are optional
and are currently not actually supported by the modules internals, so for
now you can just ignore them.
The `ctx` argument is the context that we receive in the `OnLoad` function.
The type `name` is a 9 character name in the character set that includes
from `A-Z`, `a-z`, `0-9`, plus the underscore `_` and minus `-` characters.
Note that **this name must be unique** for each data type in the Redis
ecosystem, so be creative, use both lower-case and upper case if it makes
sense, and try to use the convention of mixing the type name with the name
of the author of the module, to create a 9 character unique name.
**NOTE:** It is very important that the name is exactly 9 chars or the
registration of the type will fail. Read more to understand why.
For example if I'm building a *b-tree* data structure and my name is *antirez*
I'll call my type **btree1-az**. The name, converted to a 64 bit integer,
is stored inside the RDB file when saving the type, and will be used when the
RDB data is loaded in order to resolve what module can load the data. If Redis
finds no matching module, the integer is converted back to a name in order to
provide some clue to the user about what module is missing in order to load
the data.
The type name is also used as a reply for the `TYPE` command when called
with a key holding the registered type.
The `encver` argument is the encoding version used by the module to store data
inside the RDB file. For example I can start with an encoding version of 0,
but later when I release version 2.0 of my module, I can switch encoding to
something better. The new module will register with an encoding version of 1,
so when it saves new RDB files, the new version will be stored on disk. However
when loading RDB files, the module `rdb_load` method will be called even if
there is data found for a different encoding version (and the encoding version
is passed as argument to `rdb_load`), so that the module can still load old
RDB files.
The last argument is a structure used in order to pass the type methods to the
registration function: `rdb_load`, `rdb_save`, `aof_rewrite`, `digest` and
`free` and `mem_usage` are all callbacks with the following prototypes and uses:
typedef void *(*RedisModuleTypeLoadFunc)(RedisModuleIO *rdb, int encver);
typedef void (*RedisModuleTypeSaveFunc)(RedisModuleIO *rdb, void *value);
typedef void (*RedisModuleTypeRewriteFunc)(RedisModuleIO *aof, RedisModuleString *key, void *value);
typedef size_t (*RedisModuleTypeMemUsageFunc)(void *value);
typedef void (*RedisModuleTypeDigestFunc)(RedisModuleDigest *digest, void *value);
typedef void (*RedisModuleTypeFreeFunc)(void *value);
* `rdb_load` is called when loading data from the RDB file. It loads data in the same format as `rdb_save` produces.
* `rdb_save` is called when saving data to the RDB file.
* `aof_rewrite` is called when the AOF is being rewritten, and the module needs to tell Redis what is the sequence of commands to recreate the content of a given key.
* `digest` is called when `DEBUG DIGEST` is executed and a key holding this module type is found. Currently this is not yet implemented so the function ca be left empty.
* `mem_usage` is called when the `MEMORY` command asks for the total memory consumed by a specific key, and is used in order to get the amount of bytes used by the module value.
* `free` is called when a key with the module native type is deleted via `DEL` or in any other mean, in order to let the module reclaim the memory associated with such a value.
Ok, but *why* modules types require a 9 characters name?
---
Oh, I understand you need to understand this, so here is a very specific
explanation.
When Redis persists to RDB files, modules specific data types require to
be persisted as well. Now RDB files are sequences of key-value pairs
like the following:
[1 byte type] [key] [a type specific value]
The 1 byte type identifies strings, lists, sets, and so forth. In the case
of modules data, it is set to a special value of `module data`, but of
course this is not enough, we need the information needed to link a specific
value with a specific module type that is able to load and handle it.
So when we save a `type specific value` about a module, we prefix it with
a 64 bit integer. 64 bits is large enough to store the informations needed
in order to lookup the module that can handle that specific type, but is
short enough that we can prefix each module value we store inside the RDB
without making the final RDB file too big. At the same time, this solution
of prefixing the value with a 64 bit *signature* does not require to do
strange things like defining in the RDB header a list of modules specific
types. Everything is pretty simple.
So, what you can store in 64 bits in order to identify a given module in
a reliable way? Well if you build a character set of 64 symbols, you can
easily store 9 characters of 6 bits, and you are left with 10 bits, that
are used in order to store the *encoding version* of the type, so that
the same type can evolve in the future and provide a different and more
efficient or updated serialization format for RDB files.
So the 64 bit prefix stored before each module value is like the following:
6|6|6|6|6|6|6|6|6|10
The first 9 elements are 6-bits characters, the final 10 bits is the
encoding version.
When the RDB file is loaded back, it reads the 64 bit value, masks the final
10 bits, and searches for a matching module in the modules types cache.
When a matching one is found, the method to load the RDB file value is called
with the 10 bits encoding version as argument, so that the module knows
what version of the data layout to load, if it can support multiple versions.
Now the interesting thing about all this is that, if instead the module type
cannot be resolved, since there is no loaded module having this signature,
we can convert back the 64 bit value into a 9 characters name, and print
an error to the user that includes the module type name! So that she or he
immediately realizes what's wrong.
Setting and getting keys
---
After registering our new data type in the `RedisModule_OnLoad()` function,
we also need to be able to set Redis keys having as value our native type.
This normally happens in the context of commands that write data to a key.
The native types API allow to set and get keys to module native data types,
and to test if a given key is already associated to a value of a specific data
type.
The API uses the normal modules `RedisModule_OpenKey()` low level key access
interface in order to deal with this. This is an eaxmple of setting a
native type private data structure to a Redis key:
RedisModuleKey *key = RedisModule_OpenKey(ctx,keyname,REDISMODULE_WRITE);
struct some_private_struct *data = createMyDataStructure();
RedisModule_ModuleTypeSetValue(key,MyType,data);
The function `RedisModule_ModuleTypeSetValue()` is used with a key handle open
for writing, and gets three arguments: the key handle, the reference to the
native type, as obtained during the type registration, and finally a `void*`
pointer that contains the private data implementing the module native type.
Note that Redis has no clues at all about what your data contains. It will
just call the callbacks you provided during the method registration in order
to perform operations on the type.
Similarly we can retrieve the private data from a key using this function:
struct some_private_struct *data;
data = RedisModule_ModuleTypeGetValue(key);
We can also test for a key to have our native type as value:
if (RedisModule_ModuleTypeGetType(key) == MyType) {
/* ... do something ... */
}
However for the calls to do the right thing, we need to check if the key
is empty, if it contains a value of the right kind, and so forth. So
the idiomatic code to implement a command writing to our native type
is along these lines:
RedisModuleKey *key = RedisModule_OpenKey(ctx,argv[1],
REDISMODULE_READ|REDISMODULE_WRITE);
int type = RedisModule_KeyType(key);
if (type != REDISMODULE_KEYTYPE_EMPTY &&
RedisModule_ModuleTypeGetType(key) != MyType)
{
return RedisModule_ReplyWithError(ctx,REDISMODULE_ERRORMSG_WRONGTYPE);
}
Then if we successfully verified the key is not of the wrong type, and
we are going to write to it, we usually want to create a new data structure if
the key is empty, or retrieve the reference to the value associated to the
key if there is already one:
/* Create an empty value object if the key is currently empty. */
struct some_private_struct *data;
if (type == REDISMODULE_KEYTYPE_EMPTY) {
data = createMyDataStructure();
RedisModule_ModuleTypeSetValue(key,MyTyke,data);
} else {
data = RedisModule_ModuleTypeGetValue(key);
}
/* Do something with 'data'... */
Free method
---
As already mentioned, when Redis needs to free a key holding a native type
value, it needs help from the module in order to release the memory. This
is the reason why we pass a `free` callback during the type registration:
typedef void (*RedisModuleTypeFreeFunc)(void *value);
A trivial implementation of the free method can be something like this,
assuming our data structure is composed of a single allocation:
void MyTypeFreeCallback(void *value) {
RedisModule_Free(value);
}
However a more real world one will call some function that performs a more
complex memory reclaiming, by casting the void pointer to some structure
and freeing all the resources composing the value.
RDB load and save methods
---
The RDB saving and loading callbacks need to create (and load back) a
representation of the data type on disk. Redis offers an high level API
that can automatically store inside the RDB file the following types:
* Unsigned 64 bit integers.
* Signed 64 bit integers.
* Doubles.
* Strings.
It is up to the module to find a viable representation using the above base
types. However note that while the integer and double values are stored
and loaded in an architecture and *endianess* agnostic way, if you use
the raw string saving API to, for example, save a structure on disk, you
have to care those details yourself.
This is the list of functions performing RDB saving and loading:
void RedisModule_SaveUnsigned(RedisModuleIO *io, uint64_t value);
uint64_t RedisModule_LoadUnsigned(RedisModuleIO *io);
void RedisModule_SaveSigned(RedisModuleIO *io, int64_t value);
int64_t RedisModule_LoadSigned(RedisModuleIO *io);
void RedisModule_SaveString(RedisModuleIO *io, RedisModuleString *s);
void RedisModule_SaveStringBuffer(RedisModuleIO *io, const char *str, size_t len);
RedisModuleString *RedisModule_LoadString(RedisModuleIO *io);
char *RedisModule_LoadStringBuffer(RedisModuleIO *io, size_t *lenptr);
void RedisModule_SaveDouble(RedisModuleIO *io, double value);
double RedisModule_LoadDouble(RedisModuleIO *io);
The functions don't require any error checking from the module, that can
always assume calls succeed.
As an example, imagine I've a native type that implements an array of
double values, with the following structure:
struct double_array {
size_t count;
double *values;
};
My `rdb_save` method may look like the following:
void DoubleArrayRDBSave(RedisModuleIO *io, void *ptr) {
struct dobule_array *da = ptr;
RedisModule_SaveUnsigned(io,da->count);
for (size_t j = 0; j < da->count; j++)
RedisModule_SaveDouble(io,da->values[j]);
}
What we did was to store the number of elements followed by each double
value. So when later we'll have to load the structure in the `rdb_load`
method we'll do something like this:
void *DoubleArrayRDBLoad(RedisModuleIO *io, int encver) {
if (encver != DOUBLE_ARRAY_ENC_VER) {
/* We should actually log an error here, or try to implement
the ability to load older versions of our data structure. */
return NULL;
}
struct double_array *da;
da = RedisModule_Alloc(sizeof(*da));
da->count = RedisModule_LoadUnsigned(io);
da->values = RedisModule_Alloc(da->count * sizeof(double));
for (size_t j = 0; j < da->count; j++)
da->values = RedisModule_LoadDouble(io);
return da;
}
The load callback just reconstruct back the data structure from the data
we stored in the RDB file.
Note that while there is no error handling on the API that writes and reads
from disk, still the load callback can return NULL on errors in case what
it reads does not look correct. Redis will just panic in that case.
AOF rewriting
---
void RedisModule_EmitAOF(RedisModuleIO *io, const char *cmdname, const char *fmt, ...);
Handling multiple encodings
---
WORK IN PROGRESS
Allocating memory
---
Modules data types should try to use `RedisModule_Alloc()` functions family
in order to allocate, reallocate and release heap memory used to implement the native data structures (see the other Redis Modules documentation for detailed information).
This is not just useful in order for Redis to be able to account for the memory used by the module, but there are also more advantages:
* Redis uses the `jemalloc` allcator, that often prevents fragmentation problems that could be caused by using the libc allocator.
* When loading strings from the RDB file, the native types API is able to return strings allocated directly with `RedisModule_Alloc()`, so that the module can directly link this memory into the data structure representation, avoiding an useless copy of the data.
Even if you are using external libraries implementing your data structures, the
allocation functions provided by the module API is exactly compatible with
`malloc()`, `realloc()`, `free()` and `strdup()`, so converting the libraries
in order to use these functions should be trivial.
In case you have an external library that uses libc `malloc()`, and you want
to avoid replacing manually all the calls with the Redis Modules API calls,
an approach could be to use simple macros in order to replace the libc calls
with the Redis API calls. Something like this could work:
#define malloc RedisModule_Alloc
#define realloc RedisModule_Realloc
#define free RedisModule_Free
#define strdup RedisModule_Strdup
However take in mind that mixing libc calls with Redis API calls will result
into troubles and crashes, so if you replace calls using macros, you need to
make sure that all the calls are correctly replaced, and that the code with
the substituted calls will never, for example, attempt to call
`RedisModule_Free()` with a pointer allocated using libc `malloc()`.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment