Redis responds to an unsubscribe with one or many replies, depending
on the current subscribe state. When channels/patterns names are
provided in a command each given name will trigger a reply even if
duplicated or not subscribed to.
To know when we can return from the subscribed state we need to do
bookkeeping on pending additional unsubscribe replies, and make sure
we receive them all before switching state.
* No reuse of the previous reply callback
When multiple replies are parsed from a socket in one read
a previously found callback might get reused when the current
reply has no known callback.
This can be triggered by the added testcase which unsubscribe to
subscribed (A,B) and a non-subscribed channel (X).
Without this correction a callback for wrong channel is called.
- In 'unsubscribe B X A', B's callback is called when handling X.
- Now this is not done, i.e. there is no callback called for X.
* Re-push monitor callback for each reply
MONITORING used the same callback for all replies while parsing
multiple responses. This handling was changed to avoid calling
the wrong callback in some scenarios.
Now also change monitorings repush to work with this change.
Includes an added async monitoring testcase.
* Add test of async commands after unsubscribe
Verify that commands are handled after unsubscribing from a channel.
A command is sent before the `unsubscribe` response is received,
which currently triggers an assert in async.c:567:
`redisProcessCallbacks: Assertion `(c->flags & REDIS_SUBSCRIBED || c->flags & REDIS_MONITORING)' failed.`
* Handle async commands after an unsubscribe
When unsubscribing from the last channel we move from the `subscribe`
state to a normal state. These states uses different holders for the
command callback information.
By moving the callback info during the state change the callback order
can be maintained.
* Add test of command timeout during pubsub
A timeout of a non-subscribe command will be ignored during pubsub.
It will be handled as an idle timeout and a response is awaited for.
* Correction for command timeout during pubsub
Disconnect when a sent non-subscribe command triggers a timeout.
RESP3 allows sending commands in parallell with pubsub handling
and these commands might get responded with a REDIS_REPLY_ARRAY.
This conflicts with the pubsub response handling for RESP2 and
results in a faulty state when using RESP3.
Add functionality to keep track of PUSH/RESP3 support on the connection
and only expect the message type REDIS_REPLY_PUSH as subscribe messages
when once seen.
* Handle PING during pubsub in RESP2
* Rename invalid callback list
Some commands are valid to send during a subscribe in RESP2, and
most in RESP3. Renaming the callback list from `invalid` to `replies`
to detail this fact.
* Fix review comment
* Include `unsubscribe` as a subscribe reply in RESP3
By providing the (p)unsubscribe message via the subscribe callback,
instead of via the push callback, we get the same behavior in RESP3
as in RESP2.
* Add asynchronous test for pubsub using RESP3
The testcase will subscribe to a channel, and via a second client
a message is published to the channel. After receiving the message
the testcase will unsubscribe and disconnect.
This RESP3 testcase reuses the subscribe callback from the RESP2
testcase to make sure we have a common behavior.
When set hiredis will not automatically free replies in an async context, and the replies must be freed instead by the user.
Co-authored-by: Michael Grunder <michael.grunder@gmail.com>
Unless the callback is pushed to the list it will trigger an assert
in redisProcessCallbacks() when the response arrives.
This change let the user get an early error instead,
available in the async context directly.
Replacing the get & release functions with an initiation
function. Simplifies the code and will make sure we
run subscription callbacks in OOM scenarios.
Proper support for RESP3 PUSH messages.
By default, PUSH messages are now intercepted and the reply memory freed.
This means existing code should work unchanged when connecting to Redis
>= 6.0.0 even if `CLIENT TRACKING` were then enabled.
Additionally, we define two callbacks users can configure if they wish to handle
these messages in a custom way:
void redisPushFn(void *privdata, void *reply);
void redisAsyncPushFn(redisAsyncContext *ac, void *reply);
See #825
* Adds an indirection to every allocation/deallocation to allow users to
plug in ones of their choosing (use custom functions, jemalloc, etc).
* Gracefully handle OOM everywhere in hiredis. This should make it possible
for users of the library to have more flexibility in how they handle such situations.
* Changes `redisReaderTask->elements` from an `int` to a `long long` to prevent
a possible overflow when transferring the task elements into a `redisReply`.
* Adds a configurable `max elements` member to `redisReader` that defaults to
2^32 - 1. This can be set to "unlimited" by setting the value to zero.
Use _MSC_VER (instead of _WIN32) for things that are specific for
Visual Studio.
Also remove #include <winsock2.h> from hiredis.h, as it leaks too
many symbols and defines into the global namespace, which is
undesirable for a public interface header. Anyone who uses the
the affected parts of the hiredis API needs to include the
appropriate headers anyway in order to declare struct timeval
variables.
524: Don't pass a negative value to __redisAsyncCommand if redisFormatSdsCommandArgv fails r=badboy
525: Fix compilation on FreeBSD 10.3 with default compiler r=badboy
Given FreeBSD 10.3 with default compiler:
> $ cc -v
> FreeBSD clang version 3.4.1 (tags/RELEASE_34/dot1-final 208032) 20140512
> Target: x86_64-unknown-freebsd10.3
Defining _XOPEN_SOURCE to 600 on the fixed line leads to the following errors:
> cc -std=c99 -pedantic -c -O3 -fPIC -Wall -W -Wstrict-prototypes -Wwrite-strings -g -ggdb net.c
> net.c:435:29: error: use of undeclared identifier 'AF_LOCAL'
> if (redisCreateSocket(c,AF_LOCAL) < 0)
> ^
> net.c:460:21: error: use of undeclared identifier 'AF_LOCAL'
> sa.sun_family = AF_LOCAL;
> ^
> 2 errors generated.
>
AF_LOCAL is defined in sys/socket.h within ifdef __BSD_VISIBLE.
__BSD_VISIBLE could be defined in sys/cdefs.h, but it is never done if _XOPEN_SOURCE is defined.
So on FreeBSD _XOPEN_SOURCE shouldn't be defined.
When an asynchronous hiredis connection subscribes to a Pub/Sub channel
and gets an error, and in other related conditions, the function
redisProcessCallbacks() enters a code path where the link is
disconnected, however the function returns before freeing the allocated
reply object. This causes a memory leak. The memory leak was trivial to
trigger in Redis Sentinel, which uses hiredis, every time we tried to
subscribe to an instance that required a password, in case the Sentinel
was configured either with the wrong password or without password at
all. In this case, the -AUTH error caused the leaking code path to be
executed.
Previously, redisvAppendCommand() would return OOM even with formatting
errors. Now we use OTHER with an error string telling the user the
error was formatting related, not memory related.
This also fixes a potentially worse bug where were would pass error result
of -1 as an actual length to another function taking an unsigned length,
which would result in crash/overallocation/errors. Now for that case,
we just return an error immediately and stop processing the command.
Fixes#177
redisAsyncInitialize() can return NULL, but then we pass
the return value from redisAsyncInitialize() into something
dereferencing the return value, which can cause crashies.
[This introduces some new API functions.]
* Adds new flag to the connection context indicating SO_REUSEADDR
should be set.
* Adds max number of retries constant for when connect() hits
EADDRNOTAVAIL.
* Adds new function, redisAsyncConnectBindWithReuse(), letting
clients enable this functionality.
[Removed trailing whitespace in new header lines.]
Closes#264
