max-lag-millis parameter description is misleading when running on master

Aperta
#1,387 3 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 10 giorni

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
2/5
Tempo stimato
1-3 ore
Idoneità per principianti
48/100
Tipo di issue
Documentazione
Chiarezza
Abbastanza chiara
Stato di attività
Ferma
Stack tecnologico
aws, go, mysql

Direzione di ricerca

Inizia con il controllo del ritardo dell’heartbeat indicato in go/logic/migrator.go:504, quindi individua la documentazione relativa a --max-lag-millis e --cut-over-lock-timeout-seconds. Chiarisci che il ritardo dell’heartbeat deve rimanere al di sotto di entrambe le soglie e assicurati che le domande segnalate sul rollback e sul lock dei metadati ricevano una risposta da un maintainer.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

documentation question

This comes out of an issue we kept running into causing great pain and outages

We would run with

gh-ost --allow-on-master -assume-rbr \
-critical-load Threads_running=700 \
-critical-load-hibernate-seconds 60 \
-max-load Threads_running=100 \
-nice-ratio 0.2 \
-chunk-size 5000 \
-ask-pass \
-database=sbtest \
-table=sbtest1 \
-user sbtest \
-host=XXX \
-postpone-cut-over-flag-file /tmp/gh-ost-postpone-cut-over-flag-sbtest1.log \
-skip-foreign-key-checks -throttle-additional-flag-file /tmp/gh-ost-throttle_flag-sbtest1.log \
--alter "add column c1 int(11) NOT NULL DEFAULT '1';" \
-verbose -initially-drop-old-table -execute 2>&1 | tee /tmp/gh-ost-sbtest1.log

Then once row copy was complete, we would be in migrating stage for a long time applying binlogs. At this point heartbeat lag would be 10-30 seconds. We thought if we increased max-lag-millis from 1500 to 10000 this would give us less throttling and speed up binlog reading and applies (silly us!)

Heartbeat lag would drop below 10 seconds, we would remove the cutover file and then run into
"ERROR Timeout while waiting for events up to lock"

which made sense since 10 seconds > cutover lock timeout of 6 seconds

Our ask is that we edit the documentation to point out this important effect of the seemingly innocent parameter as evident here https://github.com/github/gh-ost/blob/master/go/logic/migrator.go#L504

if heartbeatLag > maxLagMillisecondsThrottle || heartbeatLag > cutOverLockTimeout {
				this.migrationContext.Log.Debugf("current HeartbeatLag (%.2fs) is too high, it needs to be less than both --max-lag-millis (%.2fs) and --cut-over-lock-timeout-seconds (%.2fs) to continue", heartbeatLag.Seconds(), maxLagMillisecondsThrottle.Seconds(), cutOverLockTimeout.Seconds())
				return true, nil
			}

I also have 2 questions,

  1. after this error gh-ost went back to the migrating stage, supposedly to do rollbacks which caused an availability issue and we had to kill the process. What is it rolling back, since no cutover was done, should it not simply abandon things and release locks?
  2. we obviously didn't wait for rollback to complete but should the metadata locks not released prior to the rollback and just after this error occurs? In our case both the new and old tables remained locked till we killed the process.

For context we are on AWS Aurora and the high hearbeat lag is a side effect of aurora_binlog_replication_max_yield_seconds set to non-zero

Lingua principale
Go
Stelle
13.6k
Fork
1.4k
Merge medio
9g 19h
PR unite (30g)
1

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di github/gh-ost

Tutte le issue di github/gh-ost

Issue simili

Altre issue su Go

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.