Re: Forced shutdown of data node immediately after completion of restart

[Shawn Hogan]

Thanks for the response, it's much appreciated.

Just to be clear, the option for blocking pushdown joins is at the SQL node level via "optimizer_switch=engine_condition_pushdown=off", right?

Should I make an official bug report in the bug system? Or is there some way to follow this bug so we know when it's fixed "officially" in a release?


It does seem crazy to me that something as simple as a pushdown join (enabled by default) will cause data nodes to be unable to restart. After our previous attempt to upgrade to 8.0 failed spectacularly (wrecked entire cluster, requiring restoration of backup) due to a different bug. We waited 18 months to let bugs settle and now pushdown joins make it fail. Thankfully, this wasn't as catastrophic as the previous upgrade attempt (rolled back to 7.6 and restarted data node with --initial, so no user downtime).

That being said, I love ndbcluster except for major version upgrades (they seem to be less than a 50% chance of working as expected over the last decade). I am however super appreciative of how quickly you guys are able to pinpoint whatever the problem is each time and get it sorted out.