@rjuju thanks for so swift fix of#41 . Review it shortly

Some notes about the relevance of this kind of estimation of max processes inpg_wait_sampling module.

• This value is used to allocate shared memory just for storing queryId (inproc_queryids array) as inpg_stat_kcache module. But instead ofpg_stat_kcache here we assign one slot for queryId to one slot of PGPROC therefore we have to use formula for the size ofProcGlobal->allProcs array. And the last sum itemmax_prepared_xacts might be removed as we access toproc_queryids array from regular backends and iterate over it up toProcGlobal->allProcCount index value that is without this item.
  Generally it's enough to allocateMaxBackends-sized array and access to the specific slot viabackendId-1 index value similar as it'sdone inpg_stat_kcache module.
• Starting from PG14 we might usequeryId value fromPgBackendStatus entry of sharedBackendStatusArray and don't allocate own shared memory for this purpose.@rjuju what do you think what if we rewrote the storing and extracting ofqueryId in such way in future?

And the last sum item max_prepared_xacts might be removed as we access to proc_queryids array from regular backends and iterate over it up to ProcGlobal->allProcCount index value that is without this item.

Agreed, and the code should have a comment saying that the value has to be synced with ProcGlobal->allProcCount initialization in InitProcGlobal(), so that it's clear to anyone looking at the code why this value is used.

Starting from PG14 we might use queryId value from shared BackendStatusArray and don't allocate own shared memory for queryId values, in general.@rjuju what do you think what if we rewrote the storing and extracting of queryId in such way in future?

Unfortunately it's not possible :( The community insisted to report only the top-level queryid there, for consistency with e.g. pg_stat_activity.query, even if I initially suggested to report the current query. So we, extension owners, still have to allocate and handle our own queryid array. Unfortunately there isn't even a simple way to do it only once and not once per extension.

the code should have a comment saying that the value has to be synced with ProcGlobal->allProcCount initialization in InitProcGlobal(), so that it's clear to anyone looking at the code why this value is used.

Yeah, it's a good idea to document the formula in that way.

Unfortunately it's not possible :( The community insisted to report only the top-level queryid there, for consistency with e.g. pg_stat_activity.query, even if I initially suggested to report the current query. So we, extension owners, still have to allocate and handle our own queryid array.

It's a pity.

Unfortunately there isn't even a simple way to do it only once and not once per extension.

Yes, in essence, multiple extensions have to maintain its own storage forqueryIds of currently executing queries for sharing these values between processes. And clearly, this is redundantly.

I just force-pushed the modifications we discussed. I tried to describe the problem with all consequences, so that if anyone tries to read or use the code for something else they will have all the information needed to know if that approach will work for them too.

Thanks!