room alias lookups on libera.chat timing out

[progval]

Description

I created a room yesterday, and no client can display it because getting client/v3/directory/room/ times out or crashes with a 502 error.

With curl:

$ curl https://matrix.org/_matrix/client/v3/directory/room/%23valtest8:libera.chat
{"errcode":"M_UNKNOWN","error":"Failed to fetch alias"}

With my own client, the connection times out

And from my browser (regardless of being logged in), I'm getting Cloudflare's 502 Bad Gateway page

Steps to reproduce

• join a room on a different server, with no other user from your own homeserver in the room
• leave and re-join a couple of times
• Eventually the client will be stuck

Homeserver

matrix.org

Synapse Version

1.71.0rc2 (b=matrix-org-hotfixes,39ea774cff,dirty)

Installation Method

No response

Platform

n/a

Relevant log output

n/a

Anything else that would be useful to know?

No response

[DMRobertson]

Homeserver
matrix.org

"Failed to fetch alias"

matrix.org is contacting libera.chat and asking it to resolve your alias. We are either failing to send the request or we got an error response from libera.chat.

synapse/synapse/handlers/directory.py

Lines 265 to 280 in 2bd7f3e

	try:
	fed_result: Optional[JsonDict] = await self.federation.make_query(
	destination=room_alias.domain,
	query_type="directory",
	args={"room_alias": room_alias.to_string()},
	retry_on_dns_fail=False,
	ignore_backoff=True,
	)
	except RequestSendFailed:
	raise SynapseError(502, "Failed to fetch alias")
	except CodeMessageException as e:
	logging.warning("Error retrieving alias")
	if e.code == 404:
	fed_result = None
	else:
	raise SynapseError(502, "Failed to fetch alias")

I suggest using an account on libera.chat to see if it is able to resolve the alias.

[progval]

libera.chat does not support Matrix accounts; but this request works:

$ curl https://libera.ems.host/_matrix/client/v3/directory/room/%23valtest8:libera.chat
{"room_id":"!xTdZsDoDGxqNTiMEbo:libera.chat","servers":["libera.chat","matrix.org"]}

EDIT: and over federation:

=========================================================================================
libera.chat GET https://libera.ems.host/_matrix/federation/v1/query/directory?room_alias=%23valtest8%3Alibera.chat
-----------------------------------------------------------------------------------------
200
{"room_id": "!xTdZsDoDGxqNTiMEbo:libera.chat", "servers": ["libera.chat"]}
=========================================================================================

[DMRobertson]

Matrix.org is making a similar request but not getting a response:

var/homeserver.log:2022-11-16 12:40:20,307 - synapse.http.matrixfederationclient - 672 - INFO - GET-76b04b285e9d71fe-LHR- - {GET-O-3015328} [libera.chat] Request failed: GET matrix://libera.chat/_matrix/federation/v1/query/directory?room_alias=%23valtest8%3Alibera.chat: ResponseNeverReceived:[CancelledError()]
var/homeserver.log:2022-11-16 12:40:20,307 - synapse.http.server - 108 - INFO - GET-76b04b285e9d71fe-LHR - <XForwardedForRequest at 0x7f6c060c9190 method='GET' uri='/_matrix/client/v3/directory/room/%23valtest8:libera.chat' clientproto='HTTP/1.1' site='8080'> SynapseError: 502 - Failed to fetch alias
var/homeserver.log:2022-11-16 12:40:20,308 - synapse.access.http.8080 - 460 - INFO - GET-76b04b285e9d71fe-LHR - 81.101.238.199 - 8080 - {None} Processed request: 10.004sec/0.001sec (0.000sec, 0.000sec) (0.000sec/0.000sec/0) 55B 502 "GET /_matrix/client/v3/directory/room/%23valtest8:libera.chat HTTP/1.1" "curl/7.82.0" [0 dbevts]

[progval]

@clokep federation/v1/query/directory is the one failing on libera.chat

[richvdh]

There seems to be something a bit wrong with libera.chat. I have opened #14480 to track.

[DMRobertson]

Is there anything better we can do here? The only thing I can think of is a better error message to emphasise that make it clear that the problem lies with libera.chat is, e.g. Failed to lookup #valtest8:libera.chat from libera.chat.

Edit: there doesn't seem to be a particularly useful error code defined in the spec.

[DMRobertson]

If anyone wants to take this on, the error strings seem to come from

synapse/synapse/handlers/directory.py

Lines 273 to 280 in 2bd7f3e

	except RequestSendFailed:
	raise SynapseError(502, "Failed to fetch alias")
	except CodeMessageException as e:
	logging.warning("Error retrieving alias")
	if e.code == 404:
	fed_result = None
	else:
	raise SynapseError(502, "Failed to fetch alias")

[DMRobertson]

(curl https://matrix.org/_matrix/client/v3/directory/room/%23valtest8:libera.chat is currently working. Presumably libera.chat is happier these days.)