docker image relies on init.ok, that is deleted when SST is triggered

[Miguel Angel Nieto]

Cloned from Jira PXC-701:

Our docker images use pxc-entry.sh. The code shows that it relies on a file created manually on the datadir, with name init.ok. The problem is that xtrabackup SST removes that file, so in case a node leaves and then rejoins requesting a SST, init.ok will be removed and the service won't work anymore.
It is even easy to reproduce with a single-node cluster. Stop and start the docker container all the times you want, and it will work. Remove init.ok, and it won't start anymore.
Place init.ok somewhere else, or fix xtrabackup script so it doesn't delete that file.

Reported by a customer #142420

[Kenn Takara (kenn-takara)]

Fix committed for next versions of PXC 5.6/5.7

[Travis Glenn Hansen (travisghansen)]

Thanks! Can you shed some details on the solution or point to a patch? Was this a change to the docker images/entrypoint.sh or to how xtrabackup replicates etc?

[Travis Glenn Hansen (travisghansen)]

I've noticed the 5.6 image hasn't been updated. Anyway, an answer to the previous question would probably shed some light for me on how to proceed.

[Kenn Takara (kenn-takara)]

It was a patch to the wsrep_sst_xtrabackup-v2.sh script to exclude the init.ok file. It will be released with the next 5.6/5.7 release (I don't know the exact dates when it will be released though).

in the file in scripts/wsrep_sst_xtrabackup-v2.sh
(The change is to the line with cpat)

change the line from

cpat=$(parse_cnf sst cpat '.*galera\.cache$\|.*sst_in_progress$\|.*\.sst$\|.*donor-keyring$\|.*gvwstate\.dat$\|.*grastate\.dat$\|.*\.err$\|.*\.log$\|.*RPM_UPGRADE_MARKER$\|.*RPM_UPGRADE_HISTORY$')

to this

cpat=$(parse_cnf sst cpat '.*init\.ok$\|.*galera\.cache$\|.*sst_in_progress$\|.*\.sst$\|.*donor-keyring$\|.*gvwstate\.dat$\|.*grastate\.dat$\|.*\.err$\|.*\.log$\|.*RPM_UPGRADE_MARKER$\|.*RPM_UPGRADE_HISTORY$')

Prepend ".*init\.ok$\|" to the beginning of the string after "cpat". This causes the init.ok file to be ignored when copying the files over.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PXC-1932