The idea here is that if an instance persists an event via the replication HTTP API it can return before we receive that event over replication, which can lead to races where code assumes that persisting an event immediately updates various caches (e.g. current state of the room).
Most of Synapse doesn't hit such races, so we don't do the waiting automagically, instead we do so where necessary to avoid unnecessary delays. We may decide to change our minds here if it turns out there are a lot of subtle races going on.
People probably want to look at this commit by commit.
This is primarily for allowing us to send those commands from workers, but for now simply allows us to ignore echoed RDATA/POSITION commands that we sent (we get echoes of sent commands when using redis). Currently we log a WARNING on the master process every time we receive an echoed RDATA.
The aim here is to move the command handling out of the TCP protocol classes and to also merge the client and server command handling (so that we can reuse them for redis protocol). This PR simply moves the client paths to the new `ReplicationCommandHandler`, a future PR will move the server paths too.
If there was an exception setting up one of the attributes of the Homeserver
god object, then future attempts to fetch that attribute would raise a
confusing "Cyclic dependency" error. Let's make sure that we clear the
`building` flag so that we just get the original exception.
Ref: #7169
* Remove `conn_id` usage for UserSyncCommand.
Each tcp replication connection is assigned a "conn_id", which is used
to give an ID to a remotely connected worker. In a redis world, there
will no longer be a one to one mapping between connection and instance,
so instead we need to replace such usages with an ID generated by the
remote instances and included in the replicaiton commands.
This really only effects UserSyncCommand.
* Add CLEAR_USER_SYNCS command that is sent on shutdown.
This should help with the case where a synchrotron gets restarted
gracefully, rather than rely on 5 minute timeout.
This changes the replication protocol so that the server does not send down `RDATA` for rows that happened before the client connected. Instead, the server will send a `POSITION` and clients then query the database (or master out of band) to get up to date.
The `http_proxy` and `HTTPS_PROXY` env vars can be set to a `host[:port]` value which should point to a proxy.
The address of the proxy should be excluded from IP blacklists such as the `url_preview_ip_range_blacklist`.
The proxy will then be used for
* push
* url previews
* phone-home stats
* recaptcha validation
* CAS auth validation
It will *not* be used for:
* Application Services
* Identity servers
* Outbound federation
* In worker configurations, connections from workers to masters
Fixes#4198.
We should explicitly close any db connections we open, because failing to do so
can block other transactions as per
https://github.com/matrix-org/synapse/issues/3682.
Let's also try to factor out some of the boilerplate by having server classes
define their datastore class rather than duplicating the whole of `setup`.
Server Notices use a special room which the user can't dismiss. They are
created on demand when some other bit of the code calls send_notice.
(This doesn't actually do much yet becuse we don't call send_notice anywhere)
This closes#2602
v1auth was created to account for the differences in status code between
the v1 and v2_alpha revisions of the protocol (401 vs 403 for invalid
tokens). However since those protocols were merged, this makes the r0
version/endpoint internally inconsistent, and violates the
specification for the r0 endpoint.
This might break clients that rely on this inconsistency with the
specification. This is said to affect the legacy angular reference
client. However, I feel that restoring parity with the spec is more
important. Either way, it is critical to inform developers about this
change, in case they rely on the illegal behaviour.
Signed-off-by: Adrian Tschira <nota@notafile.com>
The intention here is to split the class into the bits that can be done
on workers and the bits that have to be done on the master.
In future there will also be a class that can be run on the worker,
which will delegate work to the master when necessary.
We extract the storage-independent bits of the state group resolution out to a
separate functiom, and stick it in a new handler, in preparation for its use
from the storage layer.
Initial commit; this doesn't work yet - the LIKE filtering seems too aggressive.
It also needs _do_initial_spam to be aware of prepopulating the whole user_directory_search table with all users...
...and it needs a handle_user_signup() or something to be added so that new signups get incrementally added to the table too.
Committing it here as a WIP
Non-functional refactoring to move set_password. This means that we'll be able
to properly deactivate devices and access tokens without introducing a
dependency loop.
Non-functional refactoring to move deactivate_account. This means that we'll be
able to properly deactivate devices and access tokens without introducing a
dependency loop.
This avoids the scenario where we have four different PreviewUrlResources
configured on a single app, each of which have their own caches and cache
clearing jobs.
This was broken when device list updates were implemented, as Mailer
could no longer instantiate an AuthHandler due to a dependency on
federation sending.
Erik Johnston