writeManyInstances: improve error message if we run out of placeholders.

[sjmudd]

Description

If orchestrator uses prepared statements and uses the bulk insert options it can trigger a large bulk insert statement to be triggered which exceeds the 64k placeholders that are the limit in the MySQL server.

Related issue:

• https://bugs.mysql.com/bug.php?id=101630

I have bumped into this on several occasions and the client side workarounds while not good have to be taken. This can also adversely affect the performance of keeping a cluster up to date.

To help the orchestrator administrator understand how to resolve this I have added code to recognise the specific error and to provide some more details of why the error happened, suggesting a configuration change.

Something like:

2020-11-16T15:11:37.295707+01:00 orchestrator orchestrator[57507]: flushInstanceWriteBuffer last_seen: writeManyInstances(?,true,true): error: Error 1390: Prepared statement contains too many placeholders, len(instances): 999
, len(args): 99999.  Reduce InstanceWriteBufferSize to avoid len(args) being > 64k, a limit in the MySQL source code.

Reducing the size of InstanceWriteBufferSize or alternatively setting: "MySQLInterpolateParams": true, would work.

[sjmudd]

I see a system test failing but don't see/understand what's actually broken.

[shlomi-noach]

I see a system test failing but don't see/understand what's actually broken.

Seems like a flaky test which I should fix.