Merge pull request #4778 from filecoin-project/fix/miner-fixes

Fix scheduler lockups after storage is freed
This commit is contained in:
Łukasz Magiera 2020-11-10 19:17:10 +01:00 committed by GitHub
commit 61fa94617d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 34 additions and 19 deletions

View File

@ -139,10 +139,17 @@ func (sw *schedWorker) handleWorker() {
// wait for more tasks to be assigned by the main scheduler or for the worker
// to finish precessing a task
update, ok := sw.waitForUpdates()
update, pokeSched, ok := sw.waitForUpdates()
if !ok {
return
}
if pokeSched {
// a task has finished preparing, which can mean that we've freed some space on some worker
select {
case sched.workerChange <- struct{}{}:
default: // workerChange is buffered, and scheduling is global, so it's ok if we don't send here
}
}
if update {
break
}
@ -257,23 +264,23 @@ func (sw *schedWorker) requestWindows() bool {
return true
}
func (sw *schedWorker) waitForUpdates() (update bool, ok bool) {
func (sw *schedWorker) waitForUpdates() (update bool, sched bool, ok bool) {
select {
case <-sw.heartbeatTimer.C:
return false, true
return false, false, true
case w := <-sw.scheduledWindows:
sw.worker.wndLk.Lock()
sw.worker.activeWindows = append(sw.worker.activeWindows, w)
sw.worker.wndLk.Unlock()
return true, true
return true, false, true
case <-sw.taskDone:
log.Debugw("task done", "workerid", sw.wid)
return true, true
return true, true, true
case <-sw.sched.closing:
case <-sw.worker.closingMgr:
}
return false, false
return false, false, false
}
func (sw *schedWorker) workerCompactWindows() {

View File

@ -298,6 +298,11 @@ func (st *Local) reportHealth(ctx context.Context) {
return
}
st.reportStorage(ctx)
}
}
func (st *Local) reportStorage(ctx context.Context) {
st.localLk.RLock()
toReport := map[ID]HealthReport{}
@ -318,7 +323,6 @@ func (st *Local) reportHealth(ctx context.Context) {
}
}
}
}
func (st *Local) Reserve(ctx context.Context, sid abi.SectorID, ssize abi.SectorSize, ft storiface.SectorFileType, storageIDs storiface.SectorPaths, overheadTab map[storiface.SectorFileType]int) (func(), error) {
st.localLk.Lock()
@ -568,6 +572,8 @@ func (st *Local) removeSector(ctx context.Context, sid abi.SectorID, typ storifa
log.Errorf("removing sector (%v) from %s: %+v", sid, spath, err)
}
st.reportStorage(ctx) // report freed space
return nil
}
@ -623,6 +629,8 @@ func (st *Local) MoveStorage(ctx context.Context, s abi.SectorID, ssize abi.Sect
}
}
st.reportStorage(ctx) // report space use changes
return nil
}