* MM-30026: Use DB master when getting team members from a session
A race condition happens when the read-replica isn't updated yet
by the time a session expiry message reaches another node in the cluster.
Here is the sequence of events that can cause it:
- Server1 gets any request which has to wipe session cache.
- The SQL query is written to DB master, and a cluster message is propagated
to clear the session cache for that user.
- Now before the read-replica is updated with the master’s update,
the cluster message reaches Server2. The session cache is wiped out for that user.
- _Any random_ request for that user hits Server2. Does NOT have to be
the update team name request. The request does not find the value
in session cache, because it’s wiped off, and picks it up from the DB.
Surprise surprise, it gets the stale value. Sticks it into the cache.
By now, the read-replica is updated. But guess what, we aren’t going to
ask the DB anymore, because we have it in the cache. And the cache has the stale value.
We use a temporary approach for now by introducing a context in the DB calls so that
the useMaster information can be easily passed. And this has the added advantage of
reusing the same context for future DB calls in case it happens. And we can also
add more context keys as needed.
A proper approach needs some architectural changes. See the issue for more details.
```release-note
Fixed a bug where a session will hold on to a cached value
in an HA setup with read-replicas configured.
```
* incorporate review comments
Co-authored-by: Mattermod <mattermod@users.noreply.github.com>
* MM-30300: Add coalesce to get null content as empty string on FileInfo store
* Making explicit the fields get from the FileInfo table
* Addressing PR review comments
* Fix the team and channel filtering UT to include empty team or channel
* Fix tests that were failing before this change
Once we've activated the team/channels filter tests for PostgreSQL
and MySQL there are some tests failing so this changes fixes them
* Disable team filtering tests for DBs by now
We have a discrepancy between DB search and ES/Bleve on how to filter
teams when you have users in both teams:
- DB when filtering by one team and searching by another returns users that are in both teams
- Bleve and ES returns empty
Co-authored-by: Mario de Frutos <mario@defrutos.org>
Co-authored-by: Mattermod <mattermod@users.noreply.github.com>
* MM-29980: Optimize profilesInChannels cache to fast path
We add one more message type to the fast path- profiles in channels. There
are 2 primary reasons for this:
- This is not really a new model type, but just a map of users. And users already use
the fast path. So we can get some more gains without really investing much more code.
- A more important reason is that with the upcoming striped mutex changes, we will get
a higher throughput at the cost of a bit more CPU utilization. The reason being that
since less amount of time will be spent in lock-contention, the CPU is free to do more
stuff. So this change is to counter that increase.
As usual, this gives much better performance than the original decoder.
Micro-benchmark results
```
name old time/op new time/op delta
LRU/UserMap=new-8 16.6µs ± 3% 3.9µs ± 4% -76.15% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
LRU/UserMap=new-8 4.78kB ± 0% 2.74kB ± 0% -42.65% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
LRU/UserMap=new-8 38.0 ± 0% 30.0 ± 0% -21.05% (p=0.000 n=10+10)
```
https://mattermost.atlassian.net/browse/MM-29980
Here are some results from a load test. The comparison is done with a 2 node cluster; one running master
and one running with this patch so that it's easier to compare. The total users are 2000.
<See PR>
```release-note
NONE
```
* Fix gofmt
* Trigger CI
* Add the content field to FileInfo
* Fixing the upgrade code
* Trying to fix the text-scheme
* Fixing test-schema
* Fixing test-schema
* Moving the migration to the next version
* MM-30041: Return correct error message for user save
We were collapsing all types of user conflict into a single
error message. Fixed it by inspecting the field of the invalidError type
and returning the correct message.
https://mattermost.atlassian.net/browse/MM-30041
```release-note
NONE
```
* Fix test errors
* Fix wrong message in test when comparing error messages
Co-authored-by: Rodrigo Villablanca <villa061004@gmail.com>
* MM-29979: make websocket writes zero-alloc
Instead of allocating a new slice every time we write a message,
we create a json encoder for a byte buffer and then reset the buffer
every time we write a new message.
This allocates a buffer of a constant size per-connection, but gets rid
of a new allocation for every single write, reducing pressure on GC.
After taking a distrbution of message sizes from a load test, it was seen that
2k is a good enough buffer size within which 98.5% of messages remain.
Taking a look at the alloc profiles for (*webconn).writepump:
- master branch: 8.01% of total allocations totalling 64GB
- with this PR: 4.92% and 10GB.
https://mattermost.atlassian.net/browse/MM-29979
* Skip writing in case of an encoding error
It's was including it in the bin folder but when we package we need to include it in the dist/bin so right now the `download_mmctl` script accepts a second parameter to define the path where the binary is going to be stored
* Document extractor service
* Fixing vendor modules
* Addressing PR Review comments
* Some small simplifications
* Fixing a linter complain
* simplifying a bit the code using package variables
Co-authored-by: Mattermod <mattermod@users.noreply.github.com>
