<!-- Kamailio Pull Request Template -->
<!-- IMPORTANT: - for detailed contributing guidelines, read: https://github.com/kamailio/kamailio/blob/master/.github/CONTRIBUTING.md - pull requests must be done to master branch, unless they are backports of fixes from master branch to a stable branch - backports to stable branches must be done with 'git cherry-pick -x ...' - code is contributed under BSD for core and main components (tm, sl, auth, tls) - code is contributed GPLv2 or a compatible license for the other components - GPL code is contributed with OpenSSL licensing exception -->
#### Pre-Submission Checklist <!-- Go over all points below, and after creating the PR, tick all the checkboxes that apply --> <!-- All points should be verified, otherwise, read the CONTRIBUTING guidelines from above--> <!-- If you're unsure about any of these, don't hesitate to ask on sr-dev mailing list --> - [X] Commit message has the format required by CONTRIBUTING guide - [X] Commits are split per component (core, individual modules, libs, utils, ...) - [X] Each component has a single commit (if not, squash them into one commit) - [X] No commits to README files for modules (changes must be done to docbook files in `doc/` subfolder, the README file is autogenerated)
#### Type Of Change - [ ] Small bug fix (non-breaking change which fixes an issue) - [X] New feature (non-breaking change which adds new functionality) - [ ] Breaking change (fix or feature that would change existing functionality)
#### Checklist: <!-- Go over all points below, and after creating the PR, tick the checkboxes that apply --> - [ ] PR should be backported to stable branches - [X] Tested changes locally - [ ] Related to issue #XXXX (replace XXXX with an open issue number)
#### Description <!-- Describe your changes in detail -->
Add a mqueue parameter to control the mq_add() mode. 0 - default, no changes in behavior 1 - add unique keys, keep oldest 2 - add unique keys, keep newest You can view, comment on, or merge this pull request online at:
https://github.com/kamailio/kamailio/pull/3053
-- Commit Summary --
* mqueue: add mqueue_mode modparam
-- File Changes --
M src/modules/mqueue/doc/mqueue_admin.xml (40) M src/modules/mqueue/mqueue_api.c (44) M src/modules/mqueue/mqueue_mod.c (2)
-- Patch Links --
https://github.com/kamailio/kamailio/pull/3053.patch https://github.com/kamailio/kamailio/pull/3053.diff
Walking the list of items without acquiring the lock has the risk of race with a consumer.
Then, the module was designed to define queues with attributes specified as part of `mqueue` modparm. I understood that having a variant with dedicated modparam can help sometime, but it should not be only as dedicated modparam, therefore this has to be supported also as `mqueue` modparm attribute.
@smititelu pushed 1 commit.
8a3587004c27f1dc2aaa078c6508f5606124ccba mqueue: add mqueue_mode modparam
Updated PR. Changed the name to "mqueue_addmode" to be more explicit. Can be set either as a modparam, or when defining the mqueue, and can be different for each defined mqueue. Updated doc.
Currently under testing
what I've found out while testing is that for big mqueue(s) (> 100 000) process(es) that mq_add() call become slower, because they have to search whole mqueue, under lock.
It is expected to be slower for larger number of items in the queue. There can be some improvements by keeping a hash table with the names of the items in the key to speed up the search, but that will also add to complexity.
From my point of view, this would be a specific use case where one has to be aware of consequences to processing speed. But if it useful in some cases, then that's how it is.
In other words, I am still fine to merge it, but I think there has to be no unlock (lock release) after the if with the mode:
* https://github.com/kamailio/kamailio/pull/3053/commits/8a3587004c27f1dc2aaa0...
And the next lock_get() has to be removed. Practically, everything has to be done in a single lock_get() ... lock_release(), otherwise there is a short time frame open for races: the if block does not find the item with the name, does lock release, loses CPU for a while, and during that time another process adds an item with the same name.
The penalty is that the shm alloc and copy for the new item is done inside the lock. But probably a condition on `mh->addmode` can be leveraged to do the lock/unlock at the better place, because `mh->addmode` is set at startup and never changes, plus that `mh` does not get destroyed during runtime.
Hi, thank you for the feedback!
Yes, I was also thinking about that double lock_get()/release(). Actually in my tests I updated the code to be under a single lock_get()/release(), kept shm alloc outside the locks and shm released it when/if necessary inside lock_get()/release().
Since this new addmodes might have valid use cases in some scenarios, I will update the PR soon. In the short time frame, got no intention to update the code to using hash table, but maybe in the longer time frame will do it.
Thank you, Stefan
@smititelu pushed 1 commit.
0b3a4993ccc79e30e808bc3c172c5178700743e1 mqueue: add mqueue_mode modparam
@smititelu pushed 1 commit.
f171ed6778144dcc6da599ae57634edcbcd7057c mqueue: add mqueue_mode modparam
Updated PR to take the lock sooner and check for addmode. Release it before returning. Shm alloc new "mi" item only if need.
Merged #3053 into master.
After merging I pushed another change to use a local variable to track locking and optimize a bit what operations are done under lock. Maybe you have some time to run your tests again and see if still ok.
In the tests so far it is ok. Thank you.
-
Daniel-Constantin Mierla -
Stefan Mititelu