sigb.us/dendrite
1
0
Fork 0
Code Issues Pull requests Actions Packages Projects Releases 1 Wiki Activity

Try more servers when calling /state_ids (#2610)

Browse source 
* Try more servers when calling `/state_ids`

* More logging

* Maybe fix concurrent map write

* Revert "Maybe fix concurrent map write"

This reverts commit da0dbb8362.

* Enforce a limit of 20s per server, 5 mins total
This commit is contained in:
Neil Alexander 2022-08-03 17:37:27 +01:00 committed by GitHub
parent 9fe509b18d
commit 3bf5ae5ffe
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 18 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
roomserver/internal/input/input_missing.go
View file

@ -326,8 +326,10 @@ func (t *missingStateReq) lookupStateAfterEvent(ctx context.Context, roomVersion
return respState, true, nil return respState, true, nil
} }
logrus.WithContext(ctx).Warnf("State for event %s not available locally, falling back to federation (via %d servers)", eventID, len(t.servers))
respState, err := t.lookupStateBeforeEvent(ctx, roomVersion, roomID, eventID) respState, err := t.lookupStateBeforeEvent(ctx, roomVersion, roomID, eventID)
if err != nil { if err != nil {
logrus.WithContext(ctx).WithError(err).Errorf("Failed to look up state before event %s", eventID)
return nil, false, fmt.Errorf("t.lookupStateBeforeEvent: %w", err) return nil, false, fmt.Errorf("t.lookupStateBeforeEvent: %w", err)
} }
@ -339,6 +341,7 @@ func (t *missingStateReq) lookupStateAfterEvent(ctx context.Context, roomVersion
case nil: case nil:
// do nothing // do nothing
default: default:
logrus.WithContext(ctx).WithError(err).Errorf("Failed to look up event %s", eventID)
return nil, false, fmt.Errorf("t.lookupEvent: %w", err) return nil, false, fmt.Errorf("t.lookupEvent: %w", err)
} }
h = t.cacheAndReturn(h) h = t.cacheAndReturn(h)
@ -662,9 +665,22 @@ func (t *missingStateReq) lookupMissingStateViaStateIDs(ctx context.Context, roo
util.GetLogger(ctx).WithField("room_id", roomID).Infof("lookupMissingStateViaStateIDs %s", eventID) util.GetLogger(ctx).WithField("room_id", roomID).Infof("lookupMissingStateViaStateIDs %s", eventID)
// fetch the state event IDs at the time of the event // fetch the state event IDs at the time of the event
stateIDs, err := t.federation.LookupStateIDs(ctx, t.origin, roomID, eventID) var stateIDs gomatrixserverlib.RespStateIDs
var err error
count := 0
totalctx, totalcancel := context.WithTimeout(ctx, time.Minute*5)
for _, serverName := range t.servers {
reqctx, reqcancel := context.WithTimeout(totalctx, time.Second*20)
stateIDs, err = t.federation.LookupStateIDs(reqctx, serverName, roomID, eventID)
reqcancel()
if err == nil {
break
}
count++
}
totalcancel()
if err != nil { if err != nil {
return nil, err return nil, fmt.Errorf("t.federation.LookupStateIDs tried %d server(s), last error: %w", count, err)
} }
// work out which auth/state IDs are missing // work out which auth/state IDs are missing
wantIDs := append(stateIDs.StateEventIDs, stateIDs.AuthEventIDs...) wantIDs := append(stateIDs.StateEventIDs, stateIDs.AuthEventIDs...)