Re: Lock on ShmemVariableCache fields? - Mailing list pgsql-hackers

From Zhang Mingli
Subject Re: Lock on ShmemVariableCache fields?
Date
Msg-id 2f0598a2-4fa3-4f94-8fe5-8a481a0f8732@Spark
Whole thread Raw
In response to Lock on ShmemVariableCache fields?  (Japin Li <japinli@hotmail.com>)
Responses Re: Lock on ShmemVariableCache fields?
List pgsql-hackers
HI,

On Oct 31, 2022, 10:48 +0800, Japin Li <japinli@hotmail.com>, wrote:

Hi, hackers

The VariableCacheData says nextOid and oidCount are protected by
OidGenLock. However, we update them without holding the lock on
OidGenLock in BootStrapXLOG(). Same as nextXid, for other fields
that are protected by XidGenLock, it holds the lock, see
SetTransactionIdLimit().

void
BootStrapXLOG(void)
{
[...]

ShmemVariableCache->nextXid = checkPoint.nextXid;
ShmemVariableCache->nextOid = checkPoint.nextOid;
ShmemVariableCache->oidCount = 0;

[...]

SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);

[...]
}

I also find a similar code in StartupXLOG(). Why we don't hold the lock
on OidGenLock when updating ShmemVariableCache->nextOid and
ShmemVariableCache->oidCount?

If the lock is unnecessary, I think adding some comments is better.

--
Regrads,
Japin Li.
ChengDu WenWu Information Technology Co.,Ltd.


As its name BootStrapXLOG, it’s used in BootStrap mode to initialize the template database.
The process doesn’t speak SQL and the database is not ready.
There won’t be  concurrent access to variables.

Regards,
Zhang Mingli

pgsql-hackers by date:

Previous
From: Bharath Rupireddy
Date:
Subject: Re: Adding doubly linked list type which stores the number of items in the list
Next
From: Bharath Rupireddy
Date:
Subject: Re: Use pg_pwritev_with_retry() instead of write() in dir_open_for_write() to avoid partial writes?