Even when the integrated alerts COG jobs successfully complete, there is
a data-updates error message. It turns out this is because we didn't
increase the integrated_alerts job timeout when we added the very long
extra COG steps. So, I just added an extra 7 hours to the timeout for
IntegratedAlertsSync job.

We seem to seem to get a ThrottlingException on describe_cluster()
fairly often (often happens several time every couple of nights). This
causes the entire job process to fail (or at least not finish the
post-processing). So, I'm adding a few retries if we get a ClientError
which is a ThrottlingException on the describe_cluster() call in
check_analysis().