RedisX

A light-weight Redis client library for C/C++

CfA logo
A simple, light-weight C/C++ Redis client library.

Author: Attila Kovacs

Last Updated: 18 September 2024

Table of Contents

Introduction

RedisX is a light-weight Redis client library for C/C++. As such, it should work with Redis forks / clones like Dragonfly or Valkey also. It supports both interactive and pipelined Redis queries, managing and processing subscriptions, atomic execution blocks, and LUA scripts loading. It can be used with multiple Redis servers simultaneously also.

While there are other C/C++ Redis clients available, this one is C99 compatible, and hence can be used on older platforms also. It is also small and fast, but still capable and versatile.

Rather than providing high-level support for every possible Redis command (which would probably be impossible given the pace new commands are being introduced all the time), it provides a basic framework for synchronous and asynchronous queries, with some higher-level functions for managing key/value storage types (including hash tables), and PUB/SUB. Future releases may add further higher-level functionality based on demand for such features.

The RedisX library was created, and is maintained, by Attila Kovács at the Center for Astrophysics | Harvard & Smithsonian, and it is available through the Smithsonian/redisx repository on GitHub.

There are no official releases of RedisX yet. An initial 1.0.0 release is expected in late 2024 or early 2025. Before then the API may undergo slight changes and tweaks. Use the repository as is at your own risk for now.

Prerequisites

The Smithsonian/xchange library is both a build and a runtime dependency of RedisX.

Building RedisX

The RedisX library can be built either as a shared (libredisx.so[.1]) and as a static (libredisx.a) library, depending on what suits your needs best.

You can configure the build, either by editing config.mk or else by defining the relevant environment variables prior to invoking make. The following build variables can be configured:

  • CC: The C compiler to use (default: gcc).

  • CPPFLAGS: C preprocessor flags, such as externally defined compiler constants.

  • CFLAGS: Flags to pass onto the C compiler (default: -Os -Wall -std=c99). Note, -Iinclude will be added automatically.

  • LDFLAGS: Extra linker flags (default is not set). Note, -lm -lxchange will be added automatically.

  • BUILD_MODE: You can set it to debug to enable debugging features: it will initialize the global xDebug variable to TRUE and add -g to CFLAGS.

  • CHECKEXTRA: Extra options to pass to cppcheck for the make check target

  • XCHANGE: If the Smithsonian/xchange library is not installed on your system (e.g. under /usr) set XCHANGE to where the distribution can be found. The build will expect to find xchange.h under $(XCHANGE)/include and libxchange.so / libxchange.a under $(XCHANGE)/lib or else in the default LD_LIBRARY_PATH.

After configuring, you can simply run make, which will build the shared (lib/libredisx.so[.1]) and static (lib/libredisx.a) libraries, local HTML documentation (provided doxygen is available), and performs static analysis via the check target. Or, you may build just the components you are interested in, by specifying the desired make target(s). (You can use make help to get a summary of the available make targets).

After building the library you can install the above components to the desired locations on your system. For a system-wide install you may simply run:

  $ sudo make install

Or, to install in some other locations, you may set a prefix. For example to install under /opt instead, you can:

  $ sudo make prefix=/opt install

Linking your application against RedisX

Provided you have installed the shared (libredisx.so and libxchange.so) or static (libredisx.a and libxchange.a) libraries in a location that is in your LD_LIBRARY_PATH (e.g. in /usr/lib or /usr/local/lib) you can simply link your program using the -lredisx -lxchange flags. Your Makefile may look like:

myprog: ...
	cc -o $@ $^ $(LDFLAGS) -lredisx -lxchange

(Or, you might simply add -lredisx -lxchange to LDFLAGS and use a more standard recipe.) And, in if you installed the RedisX and/or xchange libraries elsewhere, you can simply add their location(s) to LD_LIBRARY_PATH prior to linking.

Managing Redis server connections

The library maintains up to three separate connections (channels) for each separate Redis server instance used: (1) an interactive client for sequential round-trip transactions, (2) a pipeline client for bulk queries and asynchronous background processing, and (3) a subscription client for PUB/SUB requests and notifications. The interactive client is always connected, the pipeline client is connected only if explicitly requested at the time of establishing the server connection, while the subscription client is connected only as needed.

Initializing

The first step is to create a Redis object, with the server name or IP address.

  // Configure the redis server to connect to "redis.mydomain.com".
  Redis *redis = redisxInit("redis.mydomain.com");
  if (redis == NULL) {
    // Abort: something did not got to plan...
    return;
  }

Before connecting to the Redis server, you may configure optional settings, such as the TCP port number to use (if not the default 6379), and the database authentication (if any):

  Redis *redis = ...
  
  // (optional) configure a non-standard port number
  redisxSetPort(&redis, 7089);
  
  // (optional) Configure the database user (since Redis 6.0, using ACL)
  redisxSetUser(&redis, "johndoe"); 
  
  // (optional) Configure the database password...
  redisxSetPassword(&redis, mySecretPasswordString);

You might also tweak the send/receive buffer sizes to use for clients, if you find the socket defaults sub-optimal for your application:

   // (optional) Set the TCP send/rcv buffer sizes to use if not default values.
   //            This setting applies to all new connections after...
   redisxSetTcpBuf(65536);

Optionally, you can select the database index to use now (or later, after connecting), if not the default (index 0):

  Redis *redis = ...
  
  // Select the database index 2
  redisxSelectDB(redis, 2);

Note, that you can switch the database index any time, with the caveat that it’s not possible to change it for the subscription client when there are active subscriptions.

Connecting

Once configured, you can connect to the server as:

   // Connect to Redis, including a 2nd dedicated client for pipelined requests
   int status = redisxConnect(&redis, TRUE);
   if (status != X_SUCCESS) {
      // Abort: we could not connect for some reason...
      ...
      // Clean up...
      redisxDestroy(&redis);
      ...
   }

The above will establish both an interactive connection and a pipelined connection client, for processing both synchronous and asynchronous requests (and responses).

Disconnecting

When you are done with a specific Redis server, you should disconnect from it:

  Redis *redis = ...
  
  redisxDisconnect(redis);

Connection hooks

The user of the RedisX library might want to know when connections to the server are established, or when disconnections happen, and may want to perform some configuration or clean-up accordingly. For this reason, the library provides support for connection ‘hooks’ – that is custom functions that are called in the even of connecting to or disconnecting from a Redis server.

Here is an example of a connection hook, which simply prints a message about the connection to the console.

  void my_connect_hook(Redis *redis) {
     printf("Connected to Redis server: %s\n", redis->id);
  }

And, it can be added to a Redis instance, between the redisxInit() and the redisxConnect() calls.

  Redis *redis = ...
  
  redisxAddConnectHook(redis, my_connect_hook);

The same goes for disconnect hooks, using redisxAddDisconnectHook() instead.

Simple Redis queries

Redis queries are sent as strings, according the the specification of the Redis protocol. All responses sent back by the server using the RESP protocol. Specifically, Redis uses version 2.0 of the RESP protocol (a.k.a. RESP2) by default, with optional support for the newer RESP3 introduced in Redis version 6.0. The RedisX library currently processes the standard RESP2 replies only. RESP3 support to the library may be added in the future (stay tuned…)

Interactive transactions

The simplest way for running a few Redis queries is to do it in interactive mode:

  Redis *redis = ...
  RESP *resp;     // This will be the pointer we receive to the Redis response
  int status;     // execution status to be populated.

  // Send "HGET my_table my_key" request
  resp = redisxRequest(redis, "HGET", "my_table", "my_key", NULL, &status);
  
  // Check return status...
  if (status != X_SUCCESS) {
    // Oops something went wrong...
    ...
  }
  ...

The redisxRequest() sends a command with up to three arguments. If the command takes fewer than 3 parameters, then the remaining ones must be set to NULL. This function thus offers a simple interface for running most basic sequential queries. In cases where 3 parameters are not sufficient, you may use redisxArrayRequest() instead, e.g.:

  ...
  char *args[] = { "my_table", "my_key" };  // parameters as an array...

  // Send "HGET my_table my_key" request with an array of 2 parameters...
  resp = redisxRequest(redis, "HGET", args, NULL, 2, &status);
  ...

The 4th argument in the list is an optional int[] array defining the individual string lengths of the parameters (if need be, or else readily available). Here, we used NULL instead, which will use strlen() on each supplied string-terminated parameter to determine its length automatically. Specifying the length may be necessary if the individual parameters are not 0-terminated strings.

In interactive mode, each request is sent to the Redis server, and the response is collected before the call returns with that response (or NULL if there was an error).

RESP data type

All responses coming from the Redis server are represented by a dynamically allocated RESP type (defined in redisx.h) structure.

typedef struct RESP {
  char type;                    // RESP type: RESP_ARRAY, RESP_INT ...
  int n;                        // Either the integer value of a RESP_INT response, or the 
                                // dimension of the value field.
  void *value;                  // Pointer to text (char *) content or to an array of components 
                                // (RESP **)
} RESP;

whose contents are:

RESP type Redis ID n value cast in C
RESP_ARRAY * number of RESP * pointers (RESP **)
RESP_INT : integer return value (void)
RESP_SIMPLE_STRING + string length (char *)
RESP_ERROR - string length (char *)
RESP_BULK_STRING $ string length or -1 if NULL (char *)

Each RESP has a type (e.g. RESP_SIMPLE_STRING), an integer value n, and a value pointer to further data. If the type is RESP_INT, then n represents the actual return value (and the value pointer is not used). For string type values n is the number of characters in the string value (not including termination), while for RESP_ARRAY types the value is a pointer to an embedded RESP array and n is the number of elements in that.

You may check the integrity of a RESP using redisxCheckRESP(). Since RESP data is dynamically allocated, the user is responsible for discarding them once they are no longer needed, e.g. by calling redisxDestroyRESP(). The two steps may be combined to automatically discard invalid or unexpected RESP data in a single step by calling redisxCheckDestroyRESP().

  RESP *r = ...
  
  // Let's say we expect 'r' to contain of an embedded RESP array of 3 elements... 
  int status = redisxCheckDestroyRESP(r, RESP_ARRAY, 3);
  if (status != X_SUCCESS) {
     // Oops, 'r' was either NULL, or does not contain a RESP array with 3 elements...
     ...
  }
  else {
     // Process the expected response...
     ...
     redisxDestroyRESP(r);
  }

Before destroying a RESP structure, the caller may want to dereference values within it if they are to be used as is (without making copies), e.g.:

  RESP *r = ...
  char *stringValue = NULL;   // to be extracted from 'r'
  
  // Let's say we expect 'r' to contain of a simple string response (of whatever length)
  int status = redisxCheckDestroyRESP(r, RESP_SIMPLE_STRING, 0);
  if (status != X_SUCCESS) {
    // Oops, 'r' was either NULL, or it was not a simple string type
    ...
  }
  else {
    // Set 'stringValue' and dereference the value field in the RESP so it's not 
    // destroyed with the RESP itself.
    stringValue = (char *) r->value;
    r->value = NULL;
     
    redisxDestroyRESP(r);     // The 'stringValue' is still a valid pointer after! 
  }

Accessing key / value data

Getting and setting keyed values

Key/value pairs are the bread and butter of Redis. They come in two varieties: (1) there are top-level key-value pairs, and (2) there are key-value pairs organized into hash tables, where the table name is a top-level key, but the fields in the table are not. The RedisX library offers a unified approach for dealing with key/value pairs, whether they are top level or hash-tables. Simply, a table name NULL is used to refer to top-level keys.

Retrieving individual keyed values is simple:

  Redis *redis = ...;
  int len; // Variable in which we return the length of the value or an error code 
  
  // Get the "property" field from the "system:subsystem" hash table
  char *value = redisxGetStringValue(redis, "system:subsystem", "property", &len);
  if (len < 0) {
    // Oops something went wrong.
    ...
  }
  
  ...
  
  // Discard the value once it's no longer needed.
  if(value) free(value);

The same goes for top-level keyed values, using NULL for the hash table name:

  // Get value for top-level key (not stored in hash table!)
  char *value = redisxGetStringValue(redis, NULL, "my-key", &len);

Alternatively, you can get the value as an undigested RESP, using redisxGetValue() instead, which allows you to check and inspect the response in more detail (e.g. to check for potential errors, or unexpected response types).

Setting values is straightforward also:

  Redis *redis = ...;
  
  // Set the "property" field in the "system:subsystem" hash table to -2.5
  // using the interactive client connection, without requiring confirmation. 
  int status = redisxSetValue(redis, "system:subsystem", "property", "-2.5", FALSE);
  if (status != X_SUCCESS) {
    // Oops something went wrong.
    ...
  }

It’s worth noting here, that values in Redis are always represented as strings, hence non-string data, such as floating-point values, must be converted to strings first. Additionally, the redisxSetValue() function works with 0-terminated string values only, but Redis allows storing unterminated byte sequences of known length also. If you find that you need to store an unterminated string (such as a binary sequence) as a value, you may just use the lower-level redisxArrayRequest() instead to process a Redis SET or HSET command with explicit byte-length specifications.

In the above example we have set the value using the interactive client to Redis, which means that the call will return only after confirmation or result is received back from the server. As such, a subsequent redisxGetValue() of the same table/key will be guaranteed to return the updated value always.

However, we could have set the new value asynchronously over the pipeline connection (by using TRUE as the last argument). In that case, the call will return as soon as the request was sent to Redis (but not confirmed, nor possibly transmitted yet!). As such, a subsequent redisxGetValue() on the same key/value field may race the request in transit, and may return the previous value on occasion. So, it’s important to remember that while pipelining can make setting multiple Redis fields very efficient, we have to be careful about retrieving the same values afterwards from the same program thread. (Arguably, however, there should never be a need to query values we set ourselves, since we readily know what they are.)

Finally, if you want to set values for multiple fields in a Redis hash table atomically, you may use redisxMultiSet(), which provides a high-level interface to the Redis HMSET command.

Listing and Scanning

The functions redisxGetKeys() and redisxGetTable() allow to return the set of Redis keywords or all key/value pairs in a table atomically. However, these commands can be computationally expensive for large tables and/or many top-level keywords, which means that the Redis server may block for undesirably long times while the result is computed.

This is where scanning offers a less selfish (hence much preferred) alternative. Rather than returning all the keys or key/value pairs contained in a table atomically at once, it allows to do it bit by bit with byte-sized individual transactions that are guaranteed to not block the Redis server long, so it may remain responsive to other queries also. For the caller the result is the same (notwithstanding the atomicity), except that the result is computed via a series of quick Redis queries rather than with one potentially very expensive query.

For example, to retrieve all top-level Redis keys, sorted alphabetically, using the scanning approach, you may write something like:

  Redis *redis = ...
  
  int nMatches;  // We'll return the number of matching Redis keys here...
  int status;    // We'll return the error status here...
  
  //  Return all Redis top-level keywords starting with "system:"
  char **keys = redisxScanKeys(redis, "system:*", &nMatches, &status);
  if (status != X_SUCCESS) {
    // Oops something went wrong...
    ...
  }
  
  // Use 'keys' as appropriate, possibly dereferencing values we want to
  // retain in other persistent data structures...
  ...
  
  // Once done using the 'keys' array, we should destroy it
  redisxDestroyKeys(keys, nMatches);

Or, to retrieve the values from a hash table for a set of keywords that match a glob pattern:

  ...
  
  // Scan all key/value pairs in hash table "system:subsystem"
  RedisEntry *entries = redisxScanTable(redis, "system:subsystem", "*", &nMatches, &status);
  if (status != X_SUCCESS) {
    // Oops something went wrong.
    ... 
  }
  
  // Use 'entries' as appropriate, possibly dereferencing values we want to
  // retain in other persistent data structures...
  ...
  
  // Once done using the 'keys' array, we should destroy it
  redisxDestroyEntries(entries, nMatches);

Finally, you may use redisxSetScanCount() to tune just how many results should individual scan queries should return (but only if you are really itching to tweak it). Please refer to the Redis documentation on the behavior of the SCAN and HSCAN commands to learn more.

Publish/subscribe (PUB/SUB) support

Broadcasting messages

It is simple to send messages to subscribers of a given channel:

  Redis *redis = ...

  // publish a message to the "hello_channel" subscribers.
  int status = redisxPublish(redis, "hello_channel", "Hello world!", 0);

The last argument is an optional string length, if readily available, or if sending a byte sequence that is not null-terminated. If zero is used for the length, as in the example above, it will automatically determine the length of the 0-terminated string message using strlen().

Alternatively, you may use the redisxPublishAsync() instead if you want to publish on a subscription client to which you have already have exclusive access (e.g. after an appropriate redisxLockConnected() call).

Subscriptions

Subscriptions work conceptually similarly to pipelined requests. To process incoming messages you need to first specify one or more RedisSubscriberCall functions, which will process PUB/SUB notifications automatically, in the background, as soon as they are received. Each RedisSubscriberCall can pre-filter the channels for which it receives notifications, by defining a channel stem. This way, the given processor function won’t even be invoked if a notification on a completely different channel arrives. Still, each RedisSubscriberCall implementation should further check the notifying channel name as appropriate to ensure that it is in fact qualified to deal with a given message.

Here is an example RedisSubscriberCall implementation to process messages:

  void my_event_processor(const char *pattern, const char *channel, const char *msg, long len) {
    // We'll print the message onto the console
    printf("Incoming message on channel %s: %s\n", channel, msg == NULL ? "<null>" : msg);
  }

There are some basic rules (best practices) for message processing. They should be fast, and never block for extended periods. If extensive processing is required, or may need to wait extensively for some resource or mutex locking, then its best that the processing function simply places the incoming message onto a queue, and let a separate background thread do the heavy lifting without holding up the subscription processing of other callback routines, or without losing responsiveness to other incoming messages.

Also, it is important that the call should never attempt to modify or call free() on the supplied string arguments, since that would interfere with other subscriber calls.

Once the function is defined, you can activate it via:

  Redis *redis = ...
  
  int status = redisxAddSubscriber(redis, "event:", my_event_processor);
  if (status != X_SUCCESS) {
    // Oops, something went wrong...
    ...
  }

We should also start subscribing to specific channels and/or channel patterns.

  Redis *redis = ...
  
  // We subscribe to all channels that beging with "event:"...
  int status = redisxSubscribe(redis, "event:*");
  if (status != X_SUCCESS) {
    // Oops, something went wrong...
    ...
  }

The redisxSuibscribe() function will translate to either a Redis PSUBSCRIBE or SUBSCRIBE command, depending whether the pattern argument contains globbing patterns or not (respectively).

Now, we are capturing and processing all messages published to channels whose name begins with "event:", using our custom my_event_processor function.

To end the subscription, we trace back the same steps by calling redisxUnsubscribe() to stop receiving further messages to the subscription channel or pattern, and by removing the my_event_procesor subscriber function as appropriate (provided no other subscription needs it) via redisxRemoveSubscriber().

Atomic execution blocks and LUA scripts

Sometimes you may want to execute a series of Redis command atomically, such that nothing else may alter the database while the set of commands execute, so that related values are always in a coherent state. For example, you want to set or query a collection of related variables so they change together and are reported together. You have two choices. (1) you can execute the Redis commands in an execution block, or else (2) load a LUA script onto the Redis server and call it with some parameters (possibly many times over).

Execution blocks

Execution blocks offer a fairly simple way of bunching together a set of Redis commands that need to be executed atomically. Such an execution block in RedisX may look something like:

  Redis *redis = ...;
  RESP *result;
  
  // Obtain a lock on the client on which to execute the block.
  // e.g. the interactive client channel.
  RedisClient *cl = redisxGetLockedConnectedClient(redis, REDISX_INTERACTIVE_CHANNEL);
  if (cl == NULL) {
    // Abort: we don't have exclusive access to the client
    return;
  }

  // -------------------------------------------------------------------------
  // Start an atomic execution block
  redisxStartBlockAsync(cl);
  
  // Send a number of Async requests
  redisxSendRequestAsync(cl, ...);
  ...

  // Execute the block of commands above atomically, and get the resulting RESP
  result = redisxExecBlockAsync(cl);
  // -------------------------------------------------------------------------
  
  // Release exlusive access to the client
  redisxUnlockClient(cl);
  
  // Inspect the RESP, etc...
  ...

If at any point things don’t go according to plan in the middle of the block, you can call redisAbortBlockAsync() to abort and discard all prior commands submitted in the execution block already. It is important to remember that every time you call redisxStartBlockAsync(), you must call either redisxExecBlockAsync() to execute it or else redisxAbortBlockAsync() to discard it. Failure to do so, will effectively end you up with a hung Redis client.

LUA script loading and execution

LUA scripting offers a more capable version of executing complex routines on the Redis server. LUA is a scripting language akin to python, and allows you to add extra logic, string manipulation etc. to your Redis queries. Best of all, once you upload the script to the server, it can reduce network traffic significantly by not having to repeatedly submit the same set of Redis commands every single time. LUA scripts also get executed very efficiently on the server, and produce only the result you want/need without returning unnecessary intermediates.

Assuming you have prepared your LUA script appropriately, you can upload it to the Redis server as:

  Redis *redis = ...
  char *script = ...         // The LUA script as a 0-terminated string.
  char *scriptSHA1 = NULL;   // We'll store the SHA1 sum of the script here

  // Load the script onto the Redis server
  int status = redixLoadScript(redis, script, &scriptSHA1);
  if(status != X_SUCCESS) {
    // Oops, something went wrong...
    ...
  }

Redis will refer to the script by its SHA1 sum, so it’s important keep a record of it. You’ll call the script with its SHA1 sum, a set of Redis keys the script may use, and a set of other parameters it might need.

  Redis *redis = ...
  
  char *keyArgs[] = { "my-redis-key-argument", NULL }; // NULL-terminated array of keyword arguments
  char *params[] = { "some-string", "-1.11", NULL };   // NULL-terminated array of extra parameters
  
  // Execute the script, with the specified keyword arguments and parameters
  RESP *reply = redisxRunScript(redis, scriptSHA1, keyArgs, params);

  // Check and inspect the reply
  ...

Or, you can use redisxRunScriptAsync() instead to send the request to run the script, and then collect the response later, as appropriate.

One thing to keep in mind about LUA scripts is that they are not fully persistent. They will be lost each time the Redis server is restarted.

Custom Redis functions

Functions, introduced in Redis 7, offer another evolutionary step over the LUA scripting described above. Unlike scripts, functions are persistent and they can be called by name rather than a cryptic SHA1 sum. Otherwise, they offer more or less the same functionality as scripts. RedisX does not currently have a built-in high-level support for managing and calling user-defined functions, but it is a feature that may be added in the not-too-distant future. Stay tuned.

Advanced queries and pipelining

Asynchronous client processing

Sometimes you might want to micro manage how requests are sent and responses to them are received. RedisX provides a set of asynchronous client functions that do that. (You’ve seen these already further above in the Pipelined transaction section.) These functions should be called with the specific client’s mutex locked, to ensure that other threads do not interfere with your sequence of requests and responses. E.g.:

  // Obtain the appropriate client with an exclusive lock on it.
  RedisClient *cl = redisxGetLockedConnectedClient(...);
  if (cl == NULL) {
    // Abort: no such client is connected
    return;
  }
  
  // Now send commands, and receive responses as you like using the redisx...Async() calls
  ...
  
  // When done, release the lock
  redisxUnlockClient(cl);

While you have the exclusive lock you may send any number of requests, e.g. via redisxSendRequestAsync() and/or redixSendArrayRequestAsync(). Then collect replies either with redisxReadReplyAsync() or else redisxIgnoreReplyAsync(). For example, the basic anatomy of sending a single request and then receiving a response, while we have exclusive access to the client, might look something like this:

  ...
  // Send a command to Redis
  int status = redisxSendRequestAsync(cl, ...);
  
  if(status == X_SUCCESS) {
    // Read the response
    RESP *reply = redisxReadReplyAsync(cl);
    
    // check and process the response
    if(redisxCheckRESP(reply, ...) != X_SUCCESS) {
      // Ooops, not the reply what we expected...
      ...
    } 
    else {
      // Process the response
      ...
    }
    
    // Destroy the reply
    redisxDestroyRESP(reply);
  }
  ...

For the best performance, you may want to leave the processing of the replies until after you unlock the client. I.e., you only block other threads from accessing the client while you send off the requests and collect the corresponding responses. You can then analyze the responses at your leisure outside of the mutexed section.

In some cases you may be OK with just firing off some Redis commands, without necessarily caring about responses. Rather than ignoring the replies with redisxIgnoreReplyAsync() you might call redisxSkipReplyAsync() instead before redisxSendRequestAsync() to instruct Redis to not even bother about sending a response to your request (it saves time and network bandwidth!):

 // We don't want to receive a response to our next command... 
 int status = redisxSkipReplyAsync(cl);
 
 if (status == X_SUCCESS) {
   // Now send the request...
   status = redisxSendRequest(cl, ...);
 }
 
 if (status != X_SUCCESS) {
    // Ooops, the request did not go through...
    ...
 }

Of course you can build up arbitrarily complex set of queries and deal with a set of responses in different ways. Do what works best for your application.

Pipelined transactions

Depending on round-trip times over the network, interactive queries may be suitable for running up to a few thousand queries per second. For higher throughput (up to ~1 million Redis transactions per second) you may need to access the Redis database in pipelined mode. RedisX provides a dedicated pipeline client/channel to the Redis server (provided the option to enable it has been used when redixConnect() was called).

In pipeline mode, requests are sent to the Redis server over the pipeline client in quick succession, without waiting for responses to return for each one individually. Responses are processed in the background by a designated callback function (or else discarded if no callback function has been set). This is what a callback function looks like:

  // Your own function to process responses to pipelined requests...
  void my_resp_processor(RESP *r) {
    // Do what you need to do with the asynchronous responses
    // that come from Redis to bulk requests.
    ...
  }

It is important to note that the processing function should not call free on the RESP pointer argument, but it may dereference and use parts of it as appropriate (just remember to set the bits referenced elsewhere to NULL so they do not get destroyed when the pipeline listener destroys the RESP after your function is done processing it). Before sending the pipelined requests, the user first needs to specify the function to process the responses, e.g.:

  Redis *redis = ...

  redisxSetPipelineConsumer(redis, my_resp_processor);

Request are sent via the redisxSendRequestAsync() and redisxSendArrayRequestAsync() functions. Note again, the Async naming, which indicates the asynchronous nature of this calls – and which indicates that these functions should be called with the appropriate mutex locked to prevent concurrency issues, and to maintain a predictable order (very important!) for processing the responses.

  Redis *redis = ...

  // We'll use a dedicated pipeline connection for asynchronous requests
  // This way, interactive requests can still be sent independently on the interactive
  // channel independently, if need be.
  RedisClient *pipe = redisxGetLockedConnectedClient(redis, REDISX_PIPELINE_CHANNEL);

  // Check that the client is valid...
  if (pipe == NULL) {
     // Abort: we do not appear to have an active pipeline connection...
     return;
  }

  // -------------------------------------------------------------------------
  // Submit a whole bunch of asynchronous requests, e.g. from a loop...
  for (...) {
    int status = redisxSendRequestAsync(pipe, ...);
    if (status != X_SUCCESS) {
      // Oops, that did not go through...
      ...
    }
    else {
      // We probably want to keep a record of what was requested and in what order
      // so our processing function can make sense of the reponses as they arrive
      // (in the same order...)
      ...
    }
  }
  // -------------------------------------------------------------------------
  
  // Release the exclusive lock on the pipeline channel, so
  // other threads may use it now that we sent off our requests...
  redisxUnlockClient(pipe);

It is important to remember that on the pipeline client you should never try to process responses directly from the same function from which commands are being sent. That’s what the interactive connection is for. Pipeline responses are always processed by a background thread (or, if you don’t specify your callback function they will be discarded). The only thing your callback function can count on is that the same number of responses will be received as the number of asynchronous requests that were sent out, and that the responses arrive in the same order as the order in which the requests were sent.

It is up to you and your callback function to keep track of what responses are expected and in what order. Some best practices to help deal with pipeline responses are summarized here:

  • Use redisxSkipReplyAsync() prior to sending pipeline requests for which you do not need a response. (This way your callback does not have to deal with unnecessary responses at all.

  • For requests that return a value, keep a record (in a FIFO) of the expected types and your data that depends on the content of the responses. For example, for pipelined HGET commands, your FIFO should have a record that specifies that a bulk string response is expected, and a pointer to data which is used to store the returned value – so that you pipeline response processing callback function can check that the response is the expected type (and size) and knows to assign/process the response appropriately to your application data.

  • You may insert Redis PING/ECHO commands to section your responses, or to provide directives to your pipeline response processor function. You can tag them uniquely so that the echoed responses can be parsed and interpreted by your callback function. For example, you may send a PING/ECHO commands to Redis with the tag "@my_resp_processor: START sequence A", or something else meaningful that you can uniquely distinguish from all other responses that you might receive.

RedisX optimizes the pipeline client for high throughput (bandwidth), whereas the interactive and subscription clients are optimized for low-latency, at the socket level.

Error handling

The principal error handling of RedisX is an extension of that of xchange, with further error codes defined in redisx.h. The RedisX functions that return an error status (either directly, or into the integer designated by a pointer argument), can be inspected by redisxErrorDescription(), e.g.:

  Redis *redis ...
  int status = redisxSetValue(...);
  if (status != X_SUCCESS) {
    // Ooops, something went wrong...
    fprintf(stderr, "WARNING! set value: %s", redisErrorDescription(status));
    ...
  }

In addition you can define your own handler function to deal with transmission (send/receive) errors, by defining your own RedisErrorHandler function, such as:

  void my_error_handler(Redis *redis, enum redisx_channel channel, const char *op) {
    fprintf(stderr, "ERROR! %s: Redis at %s, channel %d\n", op, redis->id, channel);
  }

Then activate it as:

  Redis *redis = ...
  
  redisxSetTransmitErrorHandler(redis, my_error_handler);

After that, every time there is an error with sending or receiving packets over the network to any of the Redis clients used, your handler will report it the way you want it.

Debug support

You can enable verbose output of the RedisX library with xSetVerbose(boolean). When enabled, it will produce status messages to stderrso you can follow what’s going on. In addition (or alternatively), you can enable debug messages with xSetDebug(boolean). When enabled, all errors encountered by the library (such as invalid arguments passed) will be printed to stderr, including call traces, so you can walk back to see where the error may have originated from. (You can also enable debug messages by default by defining the DEBUG constant for the compiler, e.g. by adding -DDEBUG to CFLAGS prior to calling make).

In addition, you can use redisxDebugTraffic(boolean) to debug low-level traffic to/from the Redis server, which prints excerpts of all incoming and outgoing messages from/to Redis to stderr.

For helping to debug your application, the xchange library provides two macros: xvprintf() and xdprintf(), for printing verbose and debug messages to stderr. Both work just like printf(), but they are conditional on verbosity being enabled via xSetVerbose(boolean) and xSetDebug(boolean), respectively. Applications using RedisX may use these macros to produce their own verbose and/or debugging outputs conditional on the same global settings.

Future plans

Some obvious ways the library could evolve and grow in the not too distant future:

  • Automated regression testing and coverage tracking.
  • Keep track of subscription patterns, and automatically resubscribe to them on reconnecting.
  • Support for the RESP3 standard and Redis HELLO.
  • Support for Redis Sentinel clients, for high-availability server configurations.
  • TLS support (perhaps…)
  • Add high-level support for managing and calling custom Redis functions.
  • Add support for CLIENT TRACKING / CLIENT CACHING.
  • Add more high-level Redis commands, e.g. for lists, streams, etc.

If you have an idea for a must have feature, please let me (Attila) know. Pull requests, for new features or fixes to existing ones, are especially welcome!

Copyright (C) 2024 Attila Kovács