To: pgsql-hackers@postgresql.org
From: Fujii Masao <masao.fujii@oss.nttdata.com>
Subject: ERROR: invalid spinlock number: 0
Message-ID: <7ef8708c-5b6b-edd3-2cf2-7783f1c7c175@oss.nttdata.com>
Date: Tue, 9 Feb 2021 23:17:04 +0900
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.14; rv:78.0)
 Gecko/20100101 Thunderbird/78.7.1
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi,

Commit 3b733fcd04 caused the buildfarm member "rorqual" to report
the following error and fail the regression test. The cause of this issue
is a bug in that commit.

     ERROR:  invalid spinlock number: 0

But while investigating the issue, I found that this error could happen
even in the current master (without commit 3b733fcd04). The error can
be easily reproduced by reading pg_stat_wal_receiver view before
walreceiver starts up, in the server built with --disable-atomics --disable-spinlocks.
Furthermore if you try to read pg_stat_wal_receiver again,
that gets stuck. This is not good.

ISTM that the commit 2c8dd05d6c caused this issue. The commit changed
pg_stat_get_wal_receiver() so that it reads "writtenUpto" by using
pg_atomic_read_u64(). But since "writtenUpto" is initialized only when
walreceiver starts up, reading "writtenUpto" before the startup of
walreceiver can cause the error.

Also pg_stat_get_wal_receiver() calls pg_atomic_read_u64() while
a spinlock is being held. Maybe this may cause the process to get stuck
in --disable-atomics case, I guess.

Thought?

Regards,

-- 
Fujii Masao
Advanced Computing Technology Center
Research and Development Headquarters
NTT DATA CORPORATION