Re: Failed transaction statistics to measure the logical replication progress

On Fri, Nov 5, 2021 at 7:11 PM osumi(dot)takamichi(at)fujitsu(dot)com
<osumi(dot)takamichi(at)fujitsu(dot)com> wrote:
>

I did a quick scan through the latest v8 patch and noticed the following things:

src/backend/postmaster/pgstat.c

(1) pgstat_recv_subworker_twophase_xact()
The copying from msg->m_gid to key.gid does not seem to be correct.
strlen() is being called on a junk value, since key.gid has not been
assigned yet.
It should be changed as follows:

BEFORE:
+ strlcpy(key.gid, msg->m_gid, strlen(key.gid));
AFTER:
+ strlcpy(key.gid, msg->m_gid, sizeof(key.gid));

(2) pgstat_get_subworker_prepared_txn()
Similar to above, strlen() usage is not correct, and should use
strlcpy() instead of memcpy().

BEFORE:
+ memcpy(key.gid, gid, strlen(key.gid));
AFTER:
+ strlcpy(key.gid, gid, sizeof(key.gid));

(3) stats_reset
Note that the "stats_reset" column has been removed from the
pg_stat_subscription_workers view in the underlying latest v20 patch.

Regards,
Greg Nancarrow
Fujitsu Australia