Re: MySQL Cluster data nodes fails with 7200: LCP fragment scan watchdog

[Mikael Ronström]

1. LCP means Local CheckPoint. It writes the database content to disk to
ensure that we can cut the REDO log. So it is an absolutely necessary part
of the NDB recovery handling.

2. No issues with config that I can see

3. There are ndb tools that prints what table id 37 is. I don't know
them without looking up the docs, but it is all in the docs.

4. Restarts with ordered index builds can potentially be speeded up by
using a config parameter that builds the indexes with more indexes in
parallel. BuildIndexThreads is the name of this config variable.

Some general thoughts that could help.

1) The issue happens when the table partition has been scanned and the final
part of it is written to disk. So issues with filesystem could be one reason.
If you have lots of disk data to write (presumably not since I saw no disk
data configs in your config) it could cause this issue if the disk is very
slow.

2) If the issue is due to a bug already fixed you can always try to
upgrade to a newer version of NDB. I didn't find any such bugs at a quick
look, but there has definitely been some work done on LCP watchdog since
7.4.10, so it is possible that it could help.