Optimize backfill receiving to have less missing prev_event thrashing (scratch)

[MadLittleMods]

Addressing:

• Optimize synapse/handlers/federation_event.py#L545(backfill receiving) to be more efficient. Calculate the stream_ordering from newest -> oldest (in the correct order) and persist in the oldest -> newest to get the least missing prev_event fetch thrashing
  • Spawned from Fix historical messages backfilling in random order on remote homeservers (MSC2716) #11114 (comment)
  • Why aren't we sorting topologically when receiving backfill events? See Fix historical messages backfilling in random order on remote homeservers (MSC2716) #11114 (comment)

-- #10737

So Synapse is fast enough to merge this MSC2716 Complement test for importing many batches, matrix-org/complement#214 (comment)

---

Complement tests: matrix-org/complement#214

Dev notes

COMPLEMENT_ALWAYS_PRINT_SERVER_LOGS=1 COMPLEMENT_DIR=../complement ./scripts-dev/complement.sh -run TestImportHistoricalMessages/parallel/Federation/Backfill_still_works_after_many_contiguous_batches_are_imported -p 1

---

Why are we seeing some of these historical events being rejected?

I think it's because of changes to the auth-event reconciliation in #12943 (comment) which I was brought in on but didn't realize the magnitude of the change since the MSC2716 tests still passed. Although not sure I realized that it actually removed one the MSC2716 tests from the Synapse code base.

event's auth_events are different to our calculated auth_events. Claimed but not calculated: [<FrozenEventV3 event_id=$iTgahP49zNUQYuUfVnC0bkfNZxX_dJqu4KXUdL_GvBY, type=m.room.power_levels, state_key=, outlier=0>, <FrozenEventV3 event_id=$ycSMCM1-gaeUIFP_oR8FflBu8DSv5hfYBOk2SbHXsik, type=m.room.create, state_key=, outlier=0>, <FrozenEventV3 event_id=$COUGQXCMRhmShggKIuD7E2PiGbOKbMeXylf2RmYByIg, type=m.room.member, state_key=@maria:hs1, outlier=1>]. Calculated but not claimed: []

While checking auth of <FrozenEventV3 event_id=$VM325DZMXTz9mjGGG6pfAIK0R6ubNdmNIf5YIMdWk3o, type=m.room.message, state_key=None, outlier=False> against room state before the event: 403: User @maria:hs1 not in room !vXsKJXhHXQtktxuNfm:hs1 (None)

Why aren't we sorting topologically when receiving backfill events?

See #11114 (comment)

How is stream_ordering given out?

stream_id_gen
stream_ordering_manager

Persisting events

See _persist_events_and_state_updates for where we normally assign stream_ordering and continue down to _persist_events_txn.

_process_pulled_events
_process_pulled_event
_process_received_pdu
_run_push_actions_and_persist_event
persist_events_and_notify
persist_events
self._event_persist_queue
_process_event_persist_queue_task
_persist_event_batch
_persist_events_and_state_updates
_persist_events_txn
_store_event_txn

---

_process_pulled_event
_process_received_pdu
_check_event_auth

Pretty print list

This one is great if you're printing a JSON-like thing:

import json

json.dumps(expected_event_order, indent=4)

This one sucks because it doesn't print the first and last items on indented new lines

import pprint

pprint.pformat(expected_event_order, indent=4)

Random

\\n["|'], 

", \n\t

Pull Request Checklist

• Pull request is based on the develop branch
• Pull request includes a changelog file. The entry should:
  • Be a short description of your change which makes sense to users. "Fixed a bug that prevented receiving messages from other servers." instead of "Moved X method from EventStore to EventWorkerStore.".
  • Use markdown where necessary, mostly for code blocks.
  • End with either a period (.) or an exclamation mark (!).
  • Start with a capital letter.
  • Feel free to credit yourself, by adding a sentence "Contributed by @github_username." or "Contributed by [Your Name]." to the end of the entry.
• Pull request includes a sign off
• Code style is correct
  (run the linters)

[MadLittleMods]

This is a nice test for figuring out the mess with historical events from MSC2716 being rejected.

It eliminates all the federation variables when trying to do the same thing in Complement. And is so much faster to iterate on. Seconds vs minutes.

[MadLittleMods]

This can work without connecting it to event_before but then we rely on maria_membership_event not being gossiped about during backfill because if it is, then it will be rejected and we can't use a rejected event to auth the following historical events.

_process_pulled_event
_process_pulled_event
_compute_event_context_with_maybe_missing_prevs
compute_event_context

The reason it works when it's not gossiped about is that _compute_event_context_with_maybe_missing_prevs fills in the state_ids_before_event and resolves the state magically for us without rejecting.

synapse/synapse/handlers/federation_event.py

Lines 971 to 1037 in 5f659d4

	logger.info(
	"Event %s is missing prev_events %s: calculating state for a "
	"backwards extremity",
	event_id,
	shortstr(missing_prevs),
	)
	# Calculate the state after each of the previous events, and
	# resolve them to find the correct state at the current event.
	try:
	# Determine whether we may be about to retrieve partial state
	# Events may be un-partial stated right after we compute the partial state
	# flag, but that's okay, as long as the flag errs on the conservative side.
	partial_state_flags = await self._store.get_partial_state_events(seen)
	partial_state = any(partial_state_flags.values())
	# Get the state of the events we know about
	ours = await self._state_storage_controller.get_state_groups_ids(
	room_id, seen, await_full_state=False
	)
	# state_maps is a list of mappings from (type, state_key) to event_id
	state_maps: List[StateMap[str]] = list(ours.values())
	# we don't need this any more, let's delete it.
	del ours
	# Ask the remote server for the states we don't
	# know about
	for p in missing_prevs:
	logger.info("Requesting state after missing prev_event %s", p)
	with nested_logging_context(p):
	# note that if any of the missing prevs share missing state or
	# auth events, the requests to fetch those events are deduped
	# by the get_pdu_cache in federation_client.
	remote_state_map = (
	await self._get_state_ids_after_missing_prev_event(
	dest, room_id, p
	)
	)
	state_maps.append(remote_state_map)
	room_version = await self._store.get_room_version_id(room_id)
	state_map = await self._state_resolution_handler.resolve_events_with_store(
	room_id,
	room_version,
	state_maps,
	event_map={event_id: event},
	state_res_store=StateResolutionStore(self._store),
	)
	except Exception:
	logger.warning(
	"Error attempting to resolve state at missing prev_events",
	exc_info=True,
	)
	raise FederationError(
	"ERROR",
	403,
	"We can't get valid state history.",
	affected=event_id,
	)
	return await self._state_handler.compute_event_context(
	event, state_ids_before_event=state_map, partial_state=partial_state
	)

I have a suspicion that we will also stop doing sort of thing since we also removed auth event resolving in #12943

[MadLittleMods]

These changes make this go from before -> after where we now connect historical batch to the historical state chain. And connect the state chain to the prev_event so everything has valid prev_events and auth_events to resolve organically from the DAG.

We do lose the benefit of removing the noise @mxid joined the room noise between each batch but we might have to solve this on the client by hiding historical state.

Before

flowchart BT
    A --- annotation1>"Note: older events are at the top"]
    subgraph live timeline
        marker1>m.room.marker] ----> B -----------------> A
    end
    
    subgraph batch0
        batch0-batch[[m.room.batch]] --> batch0-2(("2")) --> batch0-1((1)) --> batch0-0((0)) --> batch0-insertion[/m.room.insertion\]
    end
    subgraph batch1
        batch1-batch[[m.room.batch]] --> batch1-2(("2")) --> batch1-1((1)) --> batch1-0((0)) --> batch1-insertion[/m.room.insertion\]
    end
    
    subgraph batch2
        batch2-batch[[m.room.batch]] --> batch2-2(("2")) --> batch2-1((1)) --> batch2-0((0)) --> batch2-insertion[/m.room.insertion\]
    end
    batch0-insertion -.-> memberBob0(["m.room.member (bob)"]) --> memberAlice0(["m.room.member (alice)"])
    batch1-insertion -.-> memberBob1(["m.room.member (bob)"]) --> memberAlice1(["m.room.member (alice)"])
    batch2-insertion -.-> memberBob2(["m.room.member (bob)"]) --> memberAlice2(["m.room.member (alice)"])
    marker1 -.-> batch0-insertionBase
    batch0-insertionBase[/m.room.insertion\] ---------------> A
    batch0-batch -.-> batch0-insertionBase
    batch1-batch -.-> batch0-insertion
    batch2-batch -.-> batch1-insertion
    %% make the annotation links invisible
    linkStyle 0 stroke-width:2px,fill:none,stroke:none;

Loading

After

flowchart BT
    A --- annotation1>"Note: older events are at the top"]
    subgraph live timeline
        marker1>m.room.marker] ----> B -----------------> A
    end
    
    subgraph batch0
        batch0-batch[[m.room.batch]] --> batch0-2(("2")) --> batch0-1((1)) --> batch0-0((0)) --> batch0-insertion[/m.room.insertion\]
    end
    subgraph batch1
        batch1-batch[[m.room.batch]] --> batch1-2(("2")) --> batch1-1((1)) --> batch1-0((0)) --> batch1-insertion[/m.room.insertion\]
    end
    
    subgraph batch2
        batch2-batch[[m.room.batch]] --> batch2-2(("2")) --> batch2-1((1)) --> batch2-0((0)) --> batch2-insertion[/m.room.insertion\]
    end
    batch0-insertion --> memberBob0(["m.room.member (bob)"]) --> memberAlice0(["m.room.member (alice)"]) --> A
    batch1-insertion --> memberBob1(["m.room.member (bob)"]) --> memberAlice1(["m.room.member (alice)"]) --> A
    batch2-insertion --> memberBob2(["m.room.member (bob)"]) --> memberAlice2(["m.room.member (alice)"]) --> A
    marker1 -.-> batch0-insertionBase
    batch0-insertionBase[/m.room.insertion\] ---------------> A
    batch0-batch -.-> batch0-insertionBase
    batch1-batch -.-> batch0-insertion
    batch2-batch -.-> batch1-insertion
    %% make the annotation links invisible
    linkStyle 0 stroke-width:2px,fill:none,stroke:none;

Loading

[MadLittleMods]

Split out to #13971