question about pending updates in pgstat_report_inj - Mailing list pgsql-hackers

I have been reviewing how custom cumulative statistics should update their
counters, and noticed something unexpected in the injection_points example
[0].

In pgstat_report_inj(), the code updates shared_stats directly:

```
shstatent = (PgStatShared_InjectionPoint *) entry_ref->shared_stats;
statent = &shstatent->stats;

/* Update the injection point statistics */
statent->numcalls++;
```

This works because it increments shared memory directly, but it bypasses the
usual pattern where updates go into ->pending and are later flushed into
shared memory by .flush_pending_cb

I think the more appropriate way to do this is:

```
    pgstat_report_inj(const char *name)
 {
        PgStat_EntryRef *entry_ref;
-       PgStatShared_InjectionPoint *shstatent;
        PgStat_StatInjEntry *statent;

        /* leave if disabled */
@@ -174,8 +173,7 @@ pgstat_report_inj(const char *name)
        entry_ref = pgstat_prep_pending_entry(PGSTAT_KIND_INJECTION, InvalidOid,

           PGSTAT_INJ_IDX(name), NULL);

-       shstatent = (PgStatShared_InjectionPoint *) entry_ref->shared_stats;
-       statent = &shstatent->stats;
+       statent = (PgStat_StatInjEntry *) entry_ref->pending;

        /* Update the injection point statistics */
        statent->numcalls++;
```

which for example pgstat_report_subscription_error does something
similar.

Maybe I am missing something obvious for injection_points?

[0] https://www.postgresql.org/docs/18/xfunc-c.html

--
Sami Imseih
Amazon Web Services (AWS)