Horton

Re: Interrupts vs signals

Heikki Linnakangas <hlinnaka@iki.fi>

View thread View on postgr.es
From: Heikki Linnakangas <hlinnaka@iki.fi>
To: Andres Freund <andres@anarazel.de>
Cc: Thomas Munro <thomas.munro@gmail.com>, Michael Paquier <michael@paquier.xyz>, Robert Haas <robertmhaas@gmail.com>, Fujii Masao <masao.fujii@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2026-03-06T14:30:37Z
Lists: pgsql-hackers

Attachments

On 20/02/2026 16:22, Heikki Linnakangas wrote:
> On 18/02/2026 02:11, Heikki Linnakangas wrote:
>> On 14/02/2026 23:56, Andres Freund wrote:
>>> Could we have the mask of interrupts that WaitInterrupt() is waiting 
>>> for in a
>>> second variable? That way we could avoid interrupting WaitInterrupt() 
>>> when
>>> raising or sending a signal that WaitInterrupt() is not waiting for. 
>>> I think
>>> that can be one race-freely with a bit of care?
>>
>> Yeah, I thought of that, but I'm not sure what the right tradeoff here 
>> is. I doubt the spurious wakeups matter much in practice. Then again, 
>> maybe it's not much more complicated, so maybe I should try that.
>>
>> Now with this new version, the same consideration applies to the 
>> CFI_ATTENTION flag I added. We could expose a process's 
>> CheckForInterruptsMask alongside the pending interrupts, so it would 
>> be SendInterrupt()'s responsibility to check if the receiving 
>> backend's CheckForInterruptsMask includes the interrupt that's being 
>> sent. That would similarly eliminate the "false positive" 
>> ProcessInterrupts() calls from CHECK_FOR_INTERRUPTS(), by moving the 
>> logic to the senders. The SLEEPING_ON_INTERRUPTS and CFI_ATTENTION 
>> flags are quite symmetrical.
> 
> I tried that approach, exposing an "attention bitmask" where a backend 
> advertises which interrupts it's currently interested in. I think I like 
> it.
> 
> Patch attached. The relevant changes for this "attention mechanism" are 
> in the last patch, but there are some other small changes too so this 
> split into patches is a little messy. I moved the list of standard 
> interrupts to a separate header file, for example. So for reviewing, I 
> recommend reading the resulting interrupt.h and interrupt.c files after 
> applying all the patches, instead of trying to read the diff for those.

Here's another rebase, no other changes.

- Heikki

Commits

  1. Ignore SIGINT in walwriter and walsummarizer

  2. Split WaitEventSet functions to separate source file

  3. Use ModifyWaitEvent to update exit_on_postmaster_death

  4. Remove unused ShutdownLatchSupport() function

  5. Rename two functions that wake up other processes

  6. Use ProcNumbers instead of direct Latch pointers to address other procs

  7. Clean up WaitLatch calls that passed latch without WL_LATCH_SET

  8. Remove unneeded #include

  9. Remove unused latch

  10. Remove support for background workers without BGWORKER_SHMEM_ACCESS.