Off and on, we've been seeing a brace of queries fail all at once, citing table errors, key corruption, giving errors like -1 and 28, and inviting us to repair what are obviously temporary tables.
We're pretty sure that what we're seeing are bad queries (botched joins) with an ORDER BY on them eating enormous amounts of /tmp space, and squeezing everybody else for space while they consume all available room. The problem straightens itself out as soon as the runaway query crashes and releases all the space it ate.
"Well, Don't Do That" is not a bad answer, but is there any way to limit how much /tmp space any one thread is allowed to use? Any way to log the query if a temp table goes past, say, fifty million rows? It'd be nice if a botched query got killed before it has a chance to take anybody else down with it.
We're pretty sure that what we're seeing are bad queries (botched joins) with an ORDER BY on them eating enormous amounts of /tmp space, and squeezing everybody else for space while they consume all available room. The problem straightens itself out as soon as the runaway query crashes and releases all the space it ate.
"Well, Don't Do That" is not a bad answer, but is there any way to limit how much /tmp space any one thread is allowed to use? Any way to log the query if a temp table goes past, say, fifty million rows? It'd be nice if a botched query got killed before it has a chance to take anybody else down with it.