I wrote:
> Tatsuo Ishii <t-ishii@sra.co.jp> writes:
>> I have a fatal error message while connecting more than 32 users using
>> current:
>> Aug 29 11:25:18 srapc1474 postgres[12189]: [1] FATAL 1:
>> ProcGetNewSemIdAndNum: cannot allocate a free semaphore
>> rather than a more informative message:
>> Sorry, too many clients already
>> Does anynbody know what's happening?
> Ugh. I thought I'd tested that.
Indeed I had, but I'd used a low value of MaxBackends to test it.
If MaxBackends is a multiple of PROC_NSEMS_PER_SET (16) then too-many-
backends will be detected here, not later on when we try to insert our
PROC entry into the sinval array. Whoops. Fix committed.
regards, tom lane