Comment 44 for bug 1704396

After ALL THE LOGGING, I managed to trace this to the select() call used to see if we have data to read from the network. Bill Erickson noticed that there were spurious SIGALRMs arriving, interrupting the select() and killing the process, which we then both traced to the use of alarm() that tells the facet fetching code to give up on retrieving facets for its caller.

Modifying retrieve_cached_facets to disable the alarm timer after the facets become available seems to address this, so a patch should be forthcoming tomorrow, after some more testing here.