> > auto_explain automatically produces the explain output of a query that is > > running for reals. The effect is identical to running explain analyze > > except your output > here is whatever the query would produce instead of > > the explain output, which instead goes into the log. > > Thanks David. I take this to mean that auto_explain produces the same > side-effects as manually running "explain analyze" does. > > Would this warrant a documentation patch ? Like so: > > auto_explain.log_analyze causes EXPLAIN ANALYZE output, rather than just > EXPLAIN output, to be printed when an execution plan is logged. Hence > the same caveats apply for INSERT/UPDATE/DELETE queries. > > Also, doesn't this makes auto_explain.log_analyze = TRUE rather surprising as > it > can make any INSERT/UPDATE/DELETE fail when it is slow for some reason ...
Ah, wait, I think I've been a bit dense here. I thought it was a two-step process of first running any queries "normally", somehow magically noticing slow ones as per auto_explain.log_min_duration, and re-running those with EPXPLAIN ANALYZE prepended. I think I understand better now: with auto_explain ALL queries are run with EXPLAIN ANALYZE prepended BUT the output is two-fold: query results are fed into whatever wire protocol client is and EXPLAIN output is re-routed to the log. Does that sound right ? I think was misguided by psql discarding (?) query results (the rows) while displaying EXPLAIN output only. The auto_explain docs might perhaps still benefit from a more explicit hint towards write query side effects. Karsten
