Thread: the number of pending entries in GIN index with FASTUPDATE=on
Hi, Is there the way to know the number of pending entries in GIN index which was created with FASTUPDATE = on? If not, is it worth implementing the function returning that number? I sometimes would like to know that number when I measure how much pending entries affect the performance of GIN index scan and tune how frequently autovacuum should run VACUUM to clean up them. Regards, -- Fujii Masao
On Fri, Nov 2, 2012 at 12:48 PM, Fujii Masao <masao.fujii@gmail.com> wrote: > Is there the way to know the number of pending entries in GIN index which > was created with FASTUPDATE = on? If not, is it worth implementing the > function returning that number? > > I sometimes would like to know that number when I measure how much > pending entries affect the performance of GIN index scan and tune how > frequently autovacuum should run VACUUM to clean up them. Seems useful to me. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
Robert Haas <robertmhaas@gmail.com> writes: > On Fri, Nov 2, 2012 at 12:48 PM, Fujii Masao <masao.fujii@gmail.com> wrote: >> Is there the way to know the number of pending entries in GIN index which >> was created with FASTUPDATE = on? If not, is it worth implementing the >> function returning that number? > Seems useful to me. Seems like the appropriate place to expose this would be in a GIN-specific variant of contrib/pgstattuple's pgstatindex(). I seem to recall some previous discussion about how to fix the btree-centricity of that function's API, but I don't recall if we thought of a good solution. regards, tom lane
On Tue, Nov 6, 2012 at 1:01 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > Robert Haas <robertmhaas@gmail.com> writes: >> On Fri, Nov 2, 2012 at 12:48 PM, Fujii Masao <masao.fujii@gmail.com> wrote: >>> Is there the way to know the number of pending entries in GIN index which >>> was created with FASTUPDATE = on? If not, is it worth implementing the >>> function returning that number? > >> Seems useful to me. > > Seems like the appropriate place to expose this would be in a > GIN-specific variant of contrib/pgstattuple's pgstatindex(). Yeah, that seems good to me, too. Or something in pgstatindex, anyway. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
On Wed, Nov 7, 2012 at 4:34 AM, Robert Haas <robertmhaas@gmail.com> wrote: > On Tue, Nov 6, 2012 at 1:01 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: >> Robert Haas <robertmhaas@gmail.com> writes: >>> On Fri, Nov 2, 2012 at 12:48 PM, Fujii Masao <masao.fujii@gmail.com> wrote: >>>> Is there the way to know the number of pending entries in GIN index which >>>> was created with FASTUPDATE = on? If not, is it worth implementing the >>>> function returning that number? >> >>> Seems useful to me. >> >> Seems like the appropriate place to expose this would be in a >> GIN-specific variant of contrib/pgstattuple's pgstatindex(). > > Yeah, that seems good to me, too. Or something in pgstatindex, anyway. Agreed. Attached patch introduces the pgstatginindex() which now reports GIN version number, number of pages in the pending list and number of tuples in the pending list, as information about a GIN index. Regards, -- Fujii Masao
Attachment
Hello, > Agreed. Attached patch introduces the pgstatginindex() which now reports > GIN version number, number of pages in the pending list and number of > tuples in the pending list, as information about a GIN index. It seems fine on the whole, and I have some suggestions. 1. This patch applies current git head cleanly, but installation ends up with failure because of the lack of pgstattuple--1.0--1.1.sqlwhich added in Makefile. 2. I feel somewhat uneasy with size for palloc's (it's too long), and BuildTupleFromCString used instead of heap_from_tuple..But it would lead additional modification for existent simillars. You can leave that if you prefer to keep this patch smaller, but it looks to me more preferable to construct the resulttuple not via c-strings in some aspects. (*1) 3. pgstatginidex shows only version, pending_pages, and pendingtuples. Why don't you show the other properties such as entrypages, data pages, entries, and total pages as pgstatindex does? regards, -- Kyotaro Horiguchi NTT Open Source Software Center (*1) Sample diff --git a/contrib/pgstattuple/pgstatindex.c b/contrib/pgstattuple/pgstatindex.c index 8a2ae85..71c2023 100644 --- a/contrib/pgstattuple/pgstatindex.c +++ b/contrib/pgstattuple/pgstatindex.c @@ -29,2 +29,3 @@ +#include "access/htup_details.h"#include "access/gin_private.h" @@ -39,3 +40,2 @@ -extern Datum pgstatindex(PG_FUNCTION_ARGS); @@ -330,4 +330,5 @@ pgstatginindex(PG_FUNCTION_ARGS) int j; - char *values[3]; + Datum values[3]; Datum result; + bool nulls[3] = {false, false, false}; @@ -376,11 +377,6 @@ pgstatginindex(PG_FUNCTION_ARGS) j = 0; - values[j] = palloc(32); - snprintf(values[j++], 32, "%d", stats.version); - values[j] = palloc(32); - snprintf(values[j++], 32, "%u", stats.pending_pages); - values[j] = palloc(64); - snprintf(values[j++], 64, INT64_FORMAT, stats.pending_tuples); - - tuple = BuildTupleFromCStrings(TupleDescGetAttInMetadata(tupleDesc), - values); + values[j++] = Int32GetDatum(stats.version); + values[j++] = UInt32GetDatum(stats.pending_pages); + values[j++] = Int64GetDatumFast(stats.pending_tuples); + tuple = heap_form_tuple(tupleDesc, values, nulls);
On Tue, Nov 20, 2012 at 4:44 PM, Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp> wrote: > Hello, > >> Agreed. Attached patch introduces the pgstatginindex() which now reports >> GIN version number, number of pages in the pending list and number of >> tuples in the pending list, as information about a GIN index. > > It seems fine on the whole, and I have some suggestions. Thanks for the review! > 1. This patch applies current git head cleanly, but installation > ends up with failure because of the lack of > pgstattuple--1.0--1.1.sql which added in Makefile. Added pgstattuple--1.0--1.1.sql. > 2. I feel somewhat uneasy with size for palloc's (it's too long), > and BuildTupleFromCString used instead of heap_from_tuple.. But > it would lead additional modification for existent simillars. > > You can leave that if you prefer to keep this patch smaller, > but it looks to me more preferable to construct the result > tuple not via c-strings in some aspects. (*1) OK. I changed the code as you suggested. Updated version of the patch attached. > > 3. pgstatginidex shows only version, pending_pages, and > pendingtuples. Why don't you show the other properties such as > entry pages, data pages, entries, and total pages as > pgstatindex does? I didn't expose those because they are accurate as of last VACUUM. But if you think they are useful, I don't object to expose them. Regards, -- Fujii Masao
Attachment
> > 1. This patch applies current git head cleanly, but installation > > ends up with failure because of the lack of > > pgstattuple--1.0--1.1.sql which added in Makefile. > > Added pgstattuple--1.0--1.1.sql. Good. Installation completed and ALTER EXTENSION UPDATE works with that. > > 2. I feel somewhat uneasy with size for palloc's (it's too long), > > and BuildTupleFromCString used instead of heap_from_tuple.. But > > it would lead additional modification for existent simillars. > > OK. I changed the code as you suggested. Thank you. It looks simpler than the last one. (although the way differs to pgstatindex..) > > 3. pgstatginidex shows only version, pending_pages, and > > pendingtuples. Why don't you show the other properties such as > > entry pages, data pages, entries, and total pages as > > pgstatindex does? > > I didn't expose those because they are accurate as of last VACUUM. > But if you think they are useful, I don't object to expose them. Ok, my point was the apparent consistency of the functions. I don't have any distinct wish about this. I'll mark this as Ready for Committer. regards, -- Kyotaro Horiguchi NTT Open Source Software Center
On 28.11.2012 04:11, Kyotaro HORIGUCHI wrote: >>> 3. pgstatginidex shows only version, pending_pages, and >>> pendingtuples. Why don't you show the other properties such as >>> entry pages, data pages, entries, and total pages as >>> pgstatindex does? >> >> I didn't expose those because they are accurate as of last VACUUM. >> But if you think they are useful, I don't object to expose them. > > Ok, my point was the apparent consistency of the functions. I > don't have any distinct wish about this. We can always add more fields later if there's a need. > I'll mark this as Ready for Committer. Thanks, committed. - Heikki