Lists: | pgsql-hackers |
---|
From: | huaxin zhang <uwcssa(at)gmail(dot)com> |
---|---|
To: | pgsql-hackers(at)postgresql(dot)org |
Subject: | counting disk access from index seek operation -- how to? |
Date: | 2005-09-12 02:20:52 |
Message-ID: | f2f5625105091119202be1d411@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Lists: | pgsql-hackers |
I need a way to tell how many pages loaded from disk for a particular
index seek operation.
What I did is to set a global flag to true before calling the
following statement
(inside index_getnext() in "/backend/access/indexam.c")
found = DatumGetBool(FunctionCall2(&scan->fn_getnext,
PointerGetDatum(scan),
Int32GetDatum(direction)));
then for each access to disk, I increment a counter until the above call is
finished and set back the global flag.
The number of page IOs is not even matching from what I got from
"select * from pg_stat_all_indexes" --- (is there anything I need to
set in postgresql.conf?)
also, the number of paged IOs for a given index seek is always less
than the total
page loads. Is it because of the statement (following the above in
"/backend/access/indexam.c")?
"if (heap_release_fetch(scan->heapRelation, scan->xs_snapshot,
heapTuple, &scan->xs_cbuf, true,
&scan->xs_pgstat_info))"
If I am running in a single user mode, is there a way to avoid using
extra page IO in
the above statement? It seems to me the extra page IO is caused by
comparing snapshots...
thanks
From: | "Qingqing Zhou" <zhouqq(at)cs(dot)toronto(dot)edu> |
---|---|
To: | pgsql-hackers(at)postgresql(dot)org |
Subject: | Re: counting disk access from index seek operation -- how to? |
Date: | 2005-09-13 00:16:16 |
Message-ID: | dg4r73$eti$1@news.hub.org |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Lists: | pgsql-hackers |
"huaxin zhang" <uwcssa(at)gmail(dot)com> wrote
>I need a way to tell how many pages loaded from disk for a particular
> index seek operation.
By pages loaded, you mean physically or logically? In either ways, I would
suggest you to take a look at _bt_getbuf().
>
> What I did is to set a global flag to true before calling the
> following statement
> (inside index_getnext() in "/backend/access/indexam.c")
>
This is not the right place. index_getnext() returns when it find a
satisifed tuple or no match at all. Thus it may access many pages, from the
root of btree down to some leaf node.
> also, the number of paged IOs for a given index seek is always less
> than the total
> page loads. Is it because of the statement (following the above in
> "/backend/access/indexam.c")?
>
> "if (heap_release_fetch(scan->heapRelation, scan->xs_snapshot,
> heapTuple, &scan->xs_cbuf, true,
> &scan->xs_pgstat_info))"
>
This functions checks if the real data on the heap matches the information
indicated by the index, since we just save a key and pointer to the real
data on index. Thus, extra IOs may needed.
Regards,
Qingqing