DB resilience when there is a power loss

[riklempens]

We received this this issue from Ubiquiti via email on November 19 2023.

DB resilience when there is a power loss

• we have to check existence of .lock file during DB start and handle it
• we have to check existence of empty .max_series_id and remove it
• same for empty series.dat and corrupt_series.dat
• there is practically no issue with Postgres under the same circumstances
• you can periodically kill power on server/droplet and you will probably hit these issues rather soon

[riklempens]

Our response via email:

We might be able to improve DB resilience for a single node but this has never been a focus for us as SirDB is designed to run as a cluster.

If you would like to continue using SiriDB as a single node we could focus our efforts on improving this in a joint effort.

As part of our testing strategy we perform various unexpected shutdown scenario's while the database is in use. We don't see the behaviour mentioned though.

[riklempens]

Response from the customer:

If you are not able to reproduce it then I think that it makes sense to close this ticket for now. I will let you know if we will be able to reproduce it and especially if we will be able to get more info. But as I wrote only issue was existence of empty files which were blocking start of DB. There was nothing else, I mean error in log, etc.

We asked the customer if it is feasible to up the SiriDB loglevel.