Incorrect formula for SysV IPC parameters

From: Kyotaro HORIGUCHI <horiguchi(dot)kyotaro(at)lab(dot)ntt(dot)co(dot)jp>
To: pgsql-hackers(at)postgresql(dot)org
Subject: Incorrect formula for SysV IPC parameters
Date: 2016-02-03 03:51:19
Message-ID: 20160203.125119.66820697.horiguchi.kyotaro@lab.ntt.co.jp
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hello, I found that the formulas to calculate SEMMNI and SEMMNS
are incorrect in 9.2 and later.

http://www.postgresql.org/docs/9.5/static/kernel-resources.html

All of them say that the same thing as following,

| SEMMNI Maximum number of semaphore identifiers (i.e., sets)
|
| at least ceil((max_connections + autovacuum_max_workers + 4) / 16)
|
| SEMMNS Maximum number of semaphores system-wide
|
| ceil((max_connections + autovacuum_max_workers + 4) / 16) * 17
| plus room for other applications

But actually the number of semaphores PostgreSQL needs is
calculated as following in 9.4 and later.

numSemas = MaxConnections + NUM_AUXILIARY_PROCS(=4)
MaxConnections = max_connections + autovacuum_max_workers + 1 +
max_worker_processes

So, the formula for SEMMNI should be

ceil((max_connections + autovacuum_max_workers + max_worker_processes + 5) / 16)

and SEMMNS should have the same fix.

In 9.3 and 9.2, the documentation says the same thing but
actually it is calculated as following,

numSemas = MaxConnections + NUM_AUXILIARY_PROCS(=4)
MaxConnections = max_connections + autovacuum_max_workers + 1 +
GetNumShmemAttachedBgworkers()

Omitting GetNumShmemAttachedBgworkers, the actual formula is

ceil((max_connections + autovacuum_max_workers + 5) / 16)

In 9.1, NUM_AUXILIARY_PROCS is 3 so the documentations is correct.

I attached two patches for 9.2-9.3 and 9.4-9.6dev
respectively. patch command complains a bit on applying it on
9.2.

On the platforforms that doesn't have tas operation needs
additional 1024 + 64 semaphores but I understand it as out of
scope of the documentation.

One concern is that 'at least' and 'plus room for other
applications' are mixed in the table 17-1, especially for SEMMNI
and SEMMNS. It seems to me that they should be in the same
wording, but it is not an actual problem.

regards,

--
Kyotaro Horiguchi
NTT Open Source Software Center

Attachment Content-Type Size
fix_doc_sysvipc_92-93.patch text/x-patch 1.3 KB
fix_doc_sysvipc_94-96.patch text/x-patch 1.4 KB

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Tom Lane 2016-02-03 03:58:15 Re: Raising the checkpoint_timeout limit
Previous Message Corey Huinker 2016-02-03 03:42:14 Re: [POC] FETCH limited by bytes.