Make indexing of user messages tolerant to user deletion.
When creating the search document for user messages,
add a check to see if the user has been deleted in the
system. If they have been deleted abort creating the search
document.
This is to avoid rasining an error when trying to get the
user context for a deleted user.
Removed the concat to generate the uniqueid field for the popup
notifications data. The concat can't be used directly in the SQL because
the syntax changes between databases. The sql_concat helper can't be
used because it assumes all values are database columns (which they
aren't in this case).
Instead I've just removed the uniqueid field because it isn't required
for the union all to work and the field isn't being used by anything.
This should fixed the compatibility issues between databases.
The OR clause in these queries used different sets of columns to select
userid which meant that the indexes that included user id could not be
used.
This change splits the query so that each individual part can use one of
the indexes that includes a userid which speeds them up considerably.
If there was no conversation found and the method returned early, the
active transaction was not marked as committable and the error was
reported by the request shutdown handler.
Attaching a unit test for the case just because it did not seem to be
covered. I was trying to reproduce the thrown error in the unit test but
it can't be used in this case. The shutdown handler puts the warning
directly into the error_log and it is not guaranteed where such messages
go (depending on the PHP configuration). And we do not even raise it
during the unit test execution (presumably due to noise it would
produce) anyway.
I have broken up the single query that was doing multiple joins on the
messaging and user tables. It no longer does any joins and will instead
query each data set individually.
This may cost more in terms of PHP processing but it will mean that
standard database tunings should be able to handle larger data sets.
For example this function was taking a long time to run on moodle.org
and was causing the MySQL join buffer to overflow.
Matt Porritt