Re: OOM on EXPLAIN with lots of nodes

Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
> Tom Lane wrote:
>> A difficulty with either your patch or my idea is that they require adding
>> another field to ExplainState, which is an ABI break for any third-party
>> code that might be declaring variables of that struct type. That's fine
>> for HEAD but would be risky to back-patch. Any thoughts about whether we
>> can get away with that (ie, anybody have an idea if there are third-party
>> extensions that call explain.c)?

> codesearch.debian.net shows a couple of hits for ExplainState in
> multicorn (an extension for FDW from Python data sources); I didn't look
> but it seems that the FDW API is using that struct.

It is, but FDWs are not at risk here: they merely reference ExplainStates
that were allocated by core backend code. So as long as we add the new
field at the end it's not a problem for them. Problematic usage would be
like what auto_explain does:

ExplainState es;

ExplainInitState(&es);
...

In hindsight, that's a bad API and we should change it to something like

ExplainState *es = NewExplainState();

so that the sizeof the struct isn't embedded in extension code. But we
definitely can't do that in back branches.

regards, tom lane