Open fleischsalatinspace opened 2 years ago
Well there is already depends_on
in pf service declaration:
depends_on:
- pfdb
- pf-redis
- pf-socket
@fleischsalatinspace maybe we could use stop_grace_period for this?
I personally have never had any issues with it probably due to very low database usage from maria, but maybe if you have heavier db utilization it might take longer than the default 10s to flush writes, and then docker-compose sends a SIGKILL.
Adding the stop_grace_period to the compose file might be a good place to start to see if mariadb just needs more time to respond to the SIGTERM
@fleischsalatinspace still having issues with this or can we close?
Just checked in the logs this still happens an all instances with a broad range of instance sizes. I have not tried stop_grace_period
. Consulting the internet i am not sure what severity this error message actually is : https://mariadb.com/kb/en/heuristic-recovery-with-the-transaction-coordinator-log/#automatic-crash-recovery-with-the-memory-mapped-file-based-transaction-coordinator-log
As mentioned above I have not noticed any impact on data integrity :shrug:
It seems like it's normal unless you have binary logging enabled? Or am I not understanding the docs correctly.
I need do investigate this further, but it looks if using
docker-compose down
mariadb container does not get shutdown successfully and crashes instead. This results in a innodb crash recovery on every db start. Mariadb container logs look like this:I haven't noticed any impact on data integrity.
If we add correctdepends_on
in the service declarations and maybe a healthcheck for every service, this issue can be solved.