Allow profile updates to happen on workers#3659
Conversation
| ) | ||
|
|
||
|
|
||
| class ProfileStore(ProfileWorkerStore): |
There was a problem hiding this comment.
Is there not a danger of races if the *_remote_profile_cache methods are called from a worker?
There was a problem hiding this comment.
on closer inspection, it looks like efforts are made to not call those methods from a worker, in which case I think they should probably stay in ProfileStore rather than moving.
| ) | ||
|
|
||
| if self.hs.config.user_directory_search_all_users: | ||
| profile = yield self.store.get_profileinfo(target_user.localpart) |
There was a problem hiding this comment.
I'm not entirely sure I'm happy that we end up with a UserDirectoryHandler which we then carefully don't use. Could we have some sort of stub handler which just provides a notify_profile_changed and is implemented by different classes on the workers than on the master?
| @@ -51,6 +52,10 @@ def __init__(self, hs): | |||
| self._start_update_remote_profile_cache, self.PROFILE_UPDATE_MS, | |||
There was a problem hiding this comment.
it might be a good time to shift this (and the things it calls) to a MasterProfileHandler
erikjohnston
force-pushed
the
erikj/split_profiles
branch
from
c3e6857 to
782689b
Compare
Turns out that the user directory handling is fairly racey as a bunch of stuff assumes that the processing happens on master, which it doesn't when there is a synapse.app.user_dir worker. So lets just call the function directly until we actually get round to fixing it, since it doesn't make the situation any worse.
|
|
||
| class ProfileHandler(BaseHandler): | ||
| class WorkerProfileHandler(BaseHandler): | ||
| PROFILE_UPDATE_MS = 60 * 1000 |
|
|
||
|
|
||
| class ProfileHandler(BaseHandler): | ||
| class WorkerProfileHandler(BaseHandler): |
There was a problem hiding this comment.
given that this is also used as a base class on the master instance, can you either give it a different name (BaseProfileHandler?) or a docstring that explains the situation?
|
|
||
| def build_profile_handler(self): | ||
| return ProfileHandler(self) | ||
| if self.config.worker_app: |
There was a problem hiding this comment.
could we have this return WorkerProfileHandler and then override it in synapse.server.HomeServer? possibly with a comment here to say that's what's going on?
There was a problem hiding this comment.
We could, but this is how we do it for other handlers. I'm not sure I really think it'll be clearer moving it into the actual app
There was a problem hiding this comment.
hrm, ok, let's punt it for now, though I'm not sure I agree.
There was a problem hiding this comment.
Certainly something we can revisit if/when we try and clean up the worker split out
| def __init__(self, hs): | ||
| super(MasterProfileHandler, self).__init__(hs) | ||
|
|
||
| self.clock.looping_call( |
There was a problem hiding this comment.
can we assert that hs.config.worker_app is None here?
|
woo, ta |
2 participants
Based on #3632