12.03.2023 10:18, Thomas Munro wrote:
> And again:
>
> TRAP: failed Assert("PMSignalState->PMChildFlags[slot] ==
> PM_CHILD_ASSIGNED"), File: "../src/backend/storage/ipc/pmsigna...
>
> https://cirrus-ci.com/task/6558324615806976
>
https://api.cirrus-ci.com/v1/artifact/task/6558324615806976/testrun/build/testrun/pg_upgrade/002_pg_upgrade/log/002_pg_upgrade_old_node.log
>
https://api.cirrus-ci.com/v1/artifact/task/6558324615806976/crashlog/crashlog-postgres.exe_0974_2023-03-11_13-57-27-982.txt
Here we have duplicate PIDs too:
...
2023-03-11 13:57:21.277 GMT [2152][client backend] [pg_regress/union][:0] LOG:
disconnection: session time: 0:00:00.268 user=SYSTEM database=regression
host=[local]
...
2023-03-11 13:57:22.320 GMT [4340][client backend] [pg_regress/join][8/947:0]
LOG: statement: set enable_hashjoin to 0;
TRAP: failed Assert("PMSignalState->PMChildFlags[slot] == PM_CHILD_ASSIGNED"),
File: "../src/backend/storage/ipc/pmsignal.c", Line: 329, PID: 2152
And I see the following code in postmaster.c:
CleanupBackend(int pid,
int exitstatus) /* child's exit status. */
{
...
dlist_foreach_modify(iter, &BackendList)
{
Backend *bp = dlist_container(Backend, elem, iter.cur);
if (bp->pid == pid)
{
if (!bp->dead_end)
{
if (!ReleasePostmasterChildSlot(bp->child_slot))
...
so if a backend with the same PID happened to start (but not reached
InitProcess() yet), when CleanBackend() is called to clean after a just
finished backend, the slot of the starting one will be released.
I am yet to construct a reproduction of the case, but it seems to me that
the race condition is not impossible here.
Best regards,
Alexander
