Обсуждение: Unexpected expensive index scan

Поиск
Список
Период
Сортировка

Unexpected expensive index scan

От
Jake Nielsen
Дата:
I've got a query that takes a surprisingly long time to run, and I'm having a really rough time trying to figure it out.

Before I get started, here are the specifics of the situation:

Here is the table that I'm working with (apologies for spammy indices, I've been throwing shit at the wall)

                            Table "public.syncerevent"

    Column    |  Type   |                        Modifiers                         

--------------+---------+----------------------------------------------------------

 id           | bigint  | not null default nextval('syncerevent_id_seq'::regclass)

 userid       | text    | 

 event        | text    | 

 eventid      | text    | 

 originatorid | text    | 

 propogatorid | text    | 

 kwargs       | text    | 

 conflicted   | integer | 

Indexes:

    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)


To provide some context, as per the wiki, 
there are 3,290,600 rows in this table. 
It gets added to frequently, but never deleted from. 
The "kwargs" column often contains mid-size JSON strings (roughly 30K characters on average)
As of right now, the table has 53 users in it. About 20% of those have a negligible number of events, but the rest of the users have a fairly even smattering.

EXPLAIN (ANALYZE, BUFFERS) says:

                                                                          QUERY PLAN                                                                          

--------------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1218.57 rows=4000 width=615) (actual time=3352.390..3403.572 rows=4000 loops=1)

   Buffers: shared hit=120244 read=160198

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..388147.29 rows=1274560 width=615) (actual time=3352.386..3383.100 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Filter: ((propogatorid <> '"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"'::text) AND (conflicted <> 1) AND (userid = '57dc984f1c87461c0967e228'::text))

         Rows Removed by Filter: 1685801

         Buffers: shared hit=120244 read=160198

 Planning time: 0.833 ms

 Execution time: 3407.633 ms

(9 rows)


The postgres verison is: PostgreSQL 9.5.2 on x86_64-pc-linux-gnu, compiled by gcc (GCC) 4.8.2 20140120 (Red Hat 4.8.2-16), 64-bit


This query has gotten slower over time.

The postgres server is running on a db.m3.medium RDS instance on Amazon.

(3.75GB of ram)

(~3 GHz processor, single core)

I ran VACUUM, and ANALYZEd this table just prior to running the EXPLAIN command.

Here are the server settings:

 name                                   | current_setting                               | source


 

 application_name                       | psql                                          | client

 archive_command                        | /etc/rds/dbbin/pgscripts/rds_wal_archive %p   | configuration file

 archive_mode                           | on                                            | configuration file

 archive_timeout                        | 5min                                          | configuration file

 autovacuum_analyze_scale_factor        | 0.05                                          | configuration file

 autovacuum_naptime                     | 30s                                           | configuration file

 autovacuum_vacuum_scale_factor         | 0.1                                           | configuration file

 checkpoint_completion_target           | 0.9                                           | configuration file

 client_encoding                        | UTF8                                          | client

 effective_cache_size                   | 1818912kB                                     | configuration file

 fsync                                  | on                                            | configuration file

 full_page_writes                       | on                                            | configuration file

 hot_standby                            | off                                           | configuration file

 listen_addresses                       | *                                             | command line

 lo_compat_privileges                   | off                                           | configuration file

 log_checkpoints                        | on                                            | configuration file

 log_directory                          | /rdsdbdata/log/error

Sorry for the formatting, I'm not sure of the best way to format this data on a mailing list.


If it matters/interests you, here is my underlying confusion:

From some internet sleuthing, I've decided that having a table per user (which would totally make this problem a non-issue) isn't a great idea. Because there is a file per table, having a table per user would not scale. My next thought was partial indexes (which would also totally help), but since there is also a table per index, this really doesn't side-step the problem. My rough mental model says: If there exists a way that a table-per-user scheme would make this more efficient, then there should also exist an index that could achieve the same effect (or close enough to not matter). I would think that "userid = '57dc984f1c87461c0967e228'" could utilize at least one of the two indexes on the userId column, but clearly I'm not understanding something.

Any help in making this query more efficient would be greatly appreciated, and any conceptual insights would be extra awesome.

Thanks for reading.

-Jake

Re: Unexpected expensive index scan

От
Jake Nielsen
Дата:
Herp, forgot to include the query:

SELECT * FROM SyncerEvent WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"') AND conflicted != 1 AND userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;^


On Tue, Sep 27, 2016 at 5:02 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:
I've got a query that takes a surprisingly long time to run, and I'm having a really rough time trying to figure it out.

Before I get started, here are the specifics of the situation:

Here is the table that I'm working with (apologies for spammy indices, I've been throwing shit at the wall)

                            Table "public.syncerevent"

    Column    |  Type   |                        Modifiers                         

--------------+---------+----------------------------------------------------------

 id           | bigint  | not null default nextval('syncerevent_id_seq'::regclass)

 userid       | text    | 

 event        | text    | 

 eventid      | text    | 

 originatorid | text    | 

 propogatorid | text    | 

 kwargs       | text    | 

 conflicted   | integer | 

Indexes:

    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)


To provide some context, as per the wiki, 
there are 3,290,600 rows in this table. 
It gets added to frequently, but never deleted from. 
The "kwargs" column often contains mid-size JSON strings (roughly 30K characters on average)
As of right now, the table has 53 users in it. About 20% of those have a negligible number of events, but the rest of the users have a fairly even smattering.

EXPLAIN (ANALYZE, BUFFERS) says:

                                                                          QUERY PLAN                                                                          

--------------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1218.57 rows=4000 width=615) (actual time=3352.390..3403.572 rows=4000 loops=1)

   Buffers: shared hit=120244 read=160198

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..388147.29 rows=1274560 width=615) (actual time=3352.386..3383.100 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Filter: ((propogatorid <> '"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"'::text) AND (conflicted <> 1) AND (userid = '57dc984f1c87461c0967e228'::text))

         Rows Removed by Filter: 1685801

         Buffers: shared hit=120244 read=160198

 Planning time: 0.833 ms

 Execution time: 3407.633 ms

(9 rows)


The postgres verison is: PostgreSQL 9.5.2 on x86_64-pc-linux-gnu, compiled by gcc (GCC) 4.8.2 20140120 (Red Hat 4.8.2-16), 64-bit


This query has gotten slower over time.

The postgres server is running on a db.m3.medium RDS instance on Amazon.

(3.75GB of ram)

(~3 GHz processor, single core)

I ran VACUUM, and ANALYZEd this table just prior to running the EXPLAIN command.

Here are the server settings:

 name                                   | current_setting                               | source


 

 application_name                       | psql                                          | client

 archive_command                        | /etc/rds/dbbin/pgscripts/rds_wal_archive %p   | configuration file

 archive_mode                           | on                                            | configuration file

 archive_timeout                        | 5min                                          | configuration file

 autovacuum_analyze_scale_factor        | 0.05                                          | configuration file

 autovacuum_naptime                     | 30s                                           | configuration file

 autovacuum_vacuum_scale_factor         | 0.1                                           | configuration file

 checkpoint_completion_target           | 0.9                                           | configuration file

 client_encoding                        | UTF8                                          | client

 effective_cache_size                   | 1818912kB                                     | configuration file

 fsync                                  | on                                            | configuration file

 full_page_writes                       | on                                            | configuration file

 hot_standby                            | off                                           | configuration file

 listen_addresses                       | *                                             | command line

 lo_compat_privileges                   | off                                           | configuration file

 log_checkpoints                        | on                                            | configuration file

 log_directory                          | /rdsdbdata/log/error

Sorry for the formatting, I'm not sure of the best way to format this data on a mailing list.


If it matters/interests you, here is my underlying confusion:

From some internet sleuthing, I've decided that having a table per user (which would totally make this problem a non-issue) isn't a great idea. Because there is a file per table, having a table per user would not scale. My next thought was partial indexes (which would also totally help), but since there is also a table per index, this really doesn't side-step the problem. My rough mental model says: If there exists a way that a table-per-user scheme would make this more efficient, then there should also exist an index that could achieve the same effect (or close enough to not matter). I would think that "userid = '57dc984f1c87461c0967e228'" could utilize at least one of the two indexes on the userId column, but clearly I'm not understanding something.

Any help in making this query more efficient would be greatly appreciated, and any conceptual insights would be extra awesome.

Thanks for reading.

-Jake


Re: Unexpected expensive index scan

От
"Mike Sofen"
Дата:

From: Jake Nielsen    Sent: Tuesday, September 27, 2016 5:22 PM


the query

SELECT * FROM SyncerEvent WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"') AND conflicted != 1 AND userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;^

 

On Tue, Sep 27, 2016 at 5:02 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:

I've got a query that takes a surprisingly long time to run, and I'm having a really rough time trying to figure it out.

 

Before I get started, here are the specifics of the situation:

 

Here is the table that I'm working with (apologies for spammy indices, I've been throwing shit at the wall)

                            Table "public.syncerevent"

    Column    |  Type   |                        Modifiers                         

--------------+---------+----------------------------------------------------------

 id           | bigint  | not null default nextval('syncerevent_id_seq'::regclass)

 userid       | text    | 

 event        | text    | 

 eventid      | text    | 

 originatorid | text    | 

 propogatorid | text    | 

 kwargs       | text    | 

 conflicted   | integer | 

Indexes:

    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)

 

To provide some context, as per the wiki, 

there are 3,290,600 rows in this table. 

It gets added to frequently, but never deleted from. 

The "kwargs" column often contains mid-size JSON strings (roughly 30K characters on average)

As of right now, the table has 53 users in it. About 20% of those have a negligible number of events, but the rest of the users have a fairly even smattering.

 

EXPLAIN (ANALYZE, BUFFERS) says:

                                                                          QUERY PLAN                                                                          

--------------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1218.57 rows=4000 width=615) (actual time=3352.390..3403.572 rows=4000 loops=1)  Buffers: shared hit=120244 read=160198

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..388147.29 rows=1274560 width=615) (actual time=3352.386..3383.100 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Filter: ((propogatorid <> '"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"'::text) AND (conflicted <> 1) AND (userid = '57dc984f1c87461c0967e228'::text))

         Rows Removed by Filter: 1685801

         Buffers: shared hit=120244 read=160198

 Planning time: 0.833 ms

 Execution time: 3407.633 ms

(9 rows)

If it matters/interests you, here is my underlying confusion:

From some internet sleuthing, I've decided that having a table per user (which would totally make this problem a non-issue) isn't a great idea. Because there is a file per table, having a table per user would not scale. My next thought was partial indexes (which would also totally help), but since there is also a table per index, this really doesn't side-step the problem. My rough mental model says: If there exists a way that a table-per-user scheme would make this more efficient, then there should also exist an index that could achieve the same effect (or close enough to not matter). I would think that "userid = '57dc984f1c87461c0967e228'" could utilize at least one of the two indexes on the userId column, but clearly I'm not understanding something.

Any help in making this query more efficient would be greatly appreciated, and any conceptual insights would be extra awesome.

Thanks for reading.

-Jake

----------------------

 

This stands out:  WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"')

As does this from the analyze:  Rows Removed by Filter: 1685801

 

The propogaterid is practically the only column NOT indexed and it’s used in a “not in”.  It looks like it’s having to do a table scan for all the rows above the id cutoff to see if any meet the filter requirement.  “not in” can be very expensive.  An index might help on this column.  Have you tried that?

 

Your rowcounts aren’t high enough to require partitioning or any other changes to your table that I can see right now.

 

Mike Sofen  (Synthetic Genomics)

 

Re: Unexpected expensive index scan

От
Jake Nielsen
Дата:


On Tue, Sep 27, 2016 at 5:41 PM, Mike Sofen <msofen@runbox.com> wrote:

From: Jake Nielsen    Sent: Tuesday, September 27, 2016 5:22 PM


the query

SELECT * FROM SyncerEvent WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"') AND conflicted != 1 AND userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;^

 

On Tue, Sep 27, 2016 at 5:02 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:

I've got a query that takes a surprisingly long time to run, and I'm having a really rough time trying to figure it out.

 

Before I get started, here are the specifics of the situation:

 

Here is the table that I'm working with (apologies for spammy indices, I've been throwing shit at the wall)

                            Table "public.syncerevent"

    Column    |  Type   |                        Modifiers                         

--------------+---------+----------------------------------------------------------

 id           | bigint  | not null default nextval('syncerevent_id_seq'::regclass)

 userid       | text    | 

 event        | text    | 

 eventid      | text    | 

 originatorid | text    | 

 propogatorid | text    | 

 kwargs       | text    | 

 conflicted   | integer | 

Indexes:

    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)

 

To provide some context, as per the wiki, 

there are 3,290,600 rows in this table. 

It gets added to frequently, but never deleted from. 

The "kwargs" column often contains mid-size JSON strings (roughly 30K characters on average)

As of right now, the table has 53 users in it. About 20% of those have a negligible number of events, but the rest of the users have a fairly even smattering.

 

EXPLAIN (ANALYZE, BUFFERS) says:

                                                                          QUERY PLAN                                                                          

--------------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1218.57 rows=4000 width=615) (actual time=3352.390..3403.572 rows=4000 loops=1)  Buffers: shared hit=120244 read=160198

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..388147.29 rows=1274560 width=615) (actual time=3352.386..3383.100 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Filter: ((propogatorid <> '"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"'::text) AND (conflicted <> 1) AND (userid = '57dc984f1c87461c0967e228'::text))

         Rows Removed by Filter: 1685801

         Buffers: shared hit=120244 read=160198

 Planning time: 0.833 ms

 Execution time: 3407.633 ms

(9 rows)

If it matters/interests you, here is my underlying confusion:

Any help in making this query more efficient would be greatly appreciated, and any conceptual insights would be extra awesome.

Thanks for reading.

-Jake

----------------------

 

This stands out:  WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"')

As does this from the analyze:  Rows Removed by Filter: 1685801

 

The propogaterid is practically the only column NOT indexed and it’s used in a “not in”.  It looks like it’s having to do a table scan for all the rows above the id cutoff to see if any meet the filter requirement.  “not in” can be very expensive.  An index might help on this column.  Have you tried that?

 

Your rowcounts aren’t high enough to require partitioning or any other changes to your table that I can see right now.

 

Mike Sofen  (Synthetic Genomics)

 


Thanks Mike, that's true, I hadn't thought of non-indexed columns forcing a scan. Unfortunately, just to test this out, I tried pulling out the more suspect parts of the query, and it still seems to want to do an index scan:


EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERE userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;

                                                                        QUERY PLAN                                                                        

----------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1140.62 rows=4000 width=615) (actual time=2706.365..2732.308 rows=4000 loops=1)

   Buffers: shared hit=120239 read=161924

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..364982.77 rows=1280431 width=615) (actual time=2706.360..2715.514 rows=4000 loops=1)

         Filter: (userid = '57dc984f1c87461c0967e228'::text)

         Rows Removed by Filter: 1698269

         Buffers: shared hit=120239 read=161924

 Planning time: 0.131 ms

 Execution time: 2748.526 ms

(8 rows)


It definitely looks to me like it's starting at the ID = 12468 row, and just grinding up the rows. The filter is (unsurprisingly) false for most of the rows, so it ends up having to chew up half the table before it actually finds 4000 rows that match.

After creating a partial index using that userId, things go way faster. This is more-or-less what I assumed I'd get by making having that multi-column index of (userId, Id), but alas:

remoteSyncerLogistics=> CREATE INDEX sillyIndex ON syncerevent (ID) where userId = '57dc984f1c87461c0967e228';

CREATE INDEX

remoteSyncerLogistics=> EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;

                                                                  QUERY PLAN                                                                  

----------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..443.21 rows=4000 width=615) (actual time=0.074..13.349 rows=4000 loops=1)

   Buffers: shared hit=842 read=13

   ->  Index Scan using sillyindex on syncerevent  (cost=0.43..141748.41 rows=1280506 width=615) (actual time=0.071..5.372 rows=4000 loops=1)

         Buffers: shared hit=842 read=13

 Planning time: 0.245 ms

 Execution time: 25.404 ms

(6 rows)


remoteSyncerLogistics=> EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId = '57dc984f1c87461c0967e228' AND ID > 12468 ORDER BY ID LIMIT 4000;

                                                                  QUERY PLAN                                                                  

----------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..453.34 rows=4000 width=615) (actual time=0.023..13.244 rows=4000 loops=1)

   Buffers: shared hit=855

   ->  Index Scan using sillyindex on syncerevent  (cost=0.43..144420.43 rows=1275492 width=615) (actual time=0.020..5.392 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Buffers: shared hit=855

 Planning time: 0.253 ms

 Execution time: 29.371 ms

(7 rows)


Any thoughts?

-Jake

Re: Unexpected expensive index scan

От
Jake Nielsen
Дата:


On Tue, Sep 27, 2016 at 6:03 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:


On Tue, Sep 27, 2016 at 5:41 PM, Mike Sofen <msofen@runbox.com> wrote:

From: Jake Nielsen    Sent: Tuesday, September 27, 2016 5:22 PM


the query

SELECT * FROM SyncerEvent WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"') AND conflicted != 1 AND userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;^

 

On Tue, Sep 27, 2016 at 5:02 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:

I've got a query that takes a surprisingly long time to run, and I'm having a really rough time trying to figure it out.

 

Before I get started, here are the specifics of the situation:

 

Here is the table that I'm working with (apologies for spammy indices, I've been throwing shit at the wall)

                            Table "public.syncerevent"

    Column    |  Type   |                        Modifiers                         

--------------+---------+----------------------------------------------------------

 id           | bigint  | not null default nextval('syncerevent_id_seq'::regclass)

 userid       | text    | 

 event        | text    | 

 eventid      | text    | 

 originatorid | text    | 

 propogatorid | text    | 

 kwargs       | text    | 

 conflicted   | integer | 

Indexes:

    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)

 

To provide some context, as per the wiki, 

there are 3,290,600 rows in this table. 

It gets added to frequently, but never deleted from. 

The "kwargs" column often contains mid-size JSON strings (roughly 30K characters on average)

As of right now, the table has 53 users in it. About 20% of those have a negligible number of events, but the rest of the users have a fairly even smattering.

 

EXPLAIN (ANALYZE, BUFFERS) says:

                                                                          QUERY PLAN                                                                          

--------------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1218.57 rows=4000 width=615) (actual time=3352.390..3403.572 rows=4000 loops=1)  Buffers: shared hit=120244 read=160198

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..388147.29 rows=1274560 width=615) (actual time=3352.386..3383.100 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Filter: ((propogatorid <> '"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"'::text) AND (conflicted <> 1) AND (userid = '57dc984f1c87461c0967e228'::text))

         Rows Removed by Filter: 1685801

         Buffers: shared hit=120244 read=160198

 Planning time: 0.833 ms

 Execution time: 3407.633 ms

(9 rows)

If it matters/interests you, here is my underlying confusion:

From some internet sleuthing, I've decided that having a table per user (which would totally make this problem a non-issue) isn't a great idea. Because there is a file per table, having a table per user would not scale. My next thought was partial indexes (which would also totally help), but since there is also a table per index, this really doesn't side-step the problem. My rough mental model says: If there exists a way that a table-per-user scheme would make this more efficient, then there should also exist an index that could achieve the same effect (or close enough to not matter). I would think that "userid = '57dc984f1c87461c0967e228'" could utilize at least one of the two indexes on the userId column, but clearly I'm not understanding something.

Any help in making this query more efficient would be greatly appreciated, and any conceptual insights would be extra awesome.

Thanks for reading.

-Jake

----------------------

 

This stands out:  WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"')

As does this from the analyze:  Rows Removed by Filter: 1685801

 

The propogaterid is practically the only column NOT indexed and it’s used in a “not in”.  It looks like it’s having to do a table scan for all the rows above the id cutoff to see if any meet the filter requirement.  “not in” can be very expensive.  An index might help on this column.  Have you tried that?

 

Your rowcounts aren’t high enough to require partitioning or any other changes to your table that I can see right now.

 

Mike Sofen  (Synthetic Genomics)

 


Thanks Mike, that's true, I hadn't thought of non-indexed columns forcing a scan. Unfortunately, just to test this out, I tried pulling out the more suspect parts of the query, and it still seems to want to do an index scan:


EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERE userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;

                                                                        QUERY PLAN                                                                        

----------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1140.62 rows=4000 width=615) (actual time=2706.365..2732.308 rows=4000 loops=1)

   Buffers: shared hit=120239 read=161924

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..364982.77 rows=1280431 width=615) (actual time=2706.360..2715.514 rows=4000 loops=1)

         Filter: (userid = '57dc984f1c87461c0967e228'::text)

         Rows Removed by Filter: 1698269

         Buffers: shared hit=120239 read=161924

 Planning time: 0.131 ms

 Execution time: 2748.526 ms

(8 rows)


It definitely looks to me like it's starting at the ID = 12468 row, and just grinding up the rows. The filter is (unsurprisingly) false for most of the rows, so it ends up having to chew up half the table before it actually finds 4000 rows that match.

After creating a partial index using that userId, things go way faster. This is more-or-less what I assumed I'd get by making having that multi-column index of (userId, Id), but alas:

remoteSyncerLogistics=> CREATE INDEX sillyIndex ON syncerevent (ID) where userId = '57dc984f1c87461c0967e228';

CREATE INDEX

remoteSyncerLogistics=> EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;

                                                                  QUERY PLAN                                                                  

----------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..443.21 rows=4000 width=615) (actual time=0.074..13.349 rows=4000 loops=1)

   Buffers: shared hit=842 read=13

   ->  Index Scan using sillyindex on syncerevent  (cost=0.43..141748.41 rows=1280506 width=615) (actual time=0.071..5.372 rows=4000 loops=1)

         Buffers: shared hit=842 read=13

 Planning time: 0.245 ms

 Execution time: 25.404 ms

(6 rows)


remoteSyncerLogistics=> EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId = '57dc984f1c87461c0967e228' AND ID > 12468 ORDER BY ID LIMIT 4000;

                                                                  QUERY PLAN                                                                  

----------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..453.34 rows=4000 width=615) (actual time=0.023..13.244 rows=4000 loops=1)

   Buffers: shared hit=855

   ->  Index Scan using sillyindex on syncerevent  (cost=0.43..144420.43 rows=1275492 width=615) (actual time=0.020..5.392 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Buffers: shared hit=855

 Planning time: 0.253 ms

 Execution time: 29.371 ms

(7 rows)


Any thoughts?

-Jake

Hmmm, here's another unexpected piece of information:


EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERE userid = '57dc984f1c87461c0967e228';

                                                        QUERY PLAN                                                         

---------------------------------------------------------------------------------------------------------------------------

 Seq Scan on syncerevent  (cost=0.00..311251.51 rows=1248302 width=618) (actual time=0.008..4970.507 rows=1259137 loops=1)

   Filter: (userid = '57dc984f1c87461c0967e228'::text)

   Rows Removed by Filter: 2032685

   Buffers: shared hit=108601 read=161523

 Planning time: 0.092 ms

 Execution time: 7662.845 ms

(6 rows)


Looks like even just using userid='blah' doesn't actually result in the index being used, despite the fact that there are indexes on the userId column:


    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)


-Jake

Re: Unexpected expensive index scan

От
Jake Nielsen
Дата:



On Tue, Sep 27, 2016 at 6:24 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:


On Tue, Sep 27, 2016 at 6:03 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:


On Tue, Sep 27, 2016 at 5:41 PM, Mike Sofen <msofen@runbox.com> wrote:

From: Jake Nielsen    Sent: Tuesday, September 27, 2016 5:22 PM


the query

SELECT * FROM SyncerEvent WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"') AND conflicted != 1 AND userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;^

 

On Tue, Sep 27, 2016 at 5:02 PM, Jake Nielsen <jake.k.nielsen@gmail.com> wrote:

I've got a query that takes a surprisingly long time to run, and I'm having a really rough time trying to figure it out.

 

Before I get started, here are the specifics of the situation:

 

Here is the table that I'm working with (apologies for spammy indices, I've been throwing shit at the wall)

                            Table "public.syncerevent"

    Column    |  Type   |                        Modifiers                         

--------------+---------+----------------------------------------------------------

 id           | bigint  | not null default nextval('syncerevent_id_seq'::regclass)

 userid       | text    | 

 event        | text    | 

 eventid      | text    | 

 originatorid | text    | 

 propogatorid | text    | 

 kwargs       | text    | 

 conflicted   | integer | 

Indexes:

    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)

 

To provide some context, as per the wiki, 

there are 3,290,600 rows in this table. 

It gets added to frequently, but never deleted from. 

The "kwargs" column often contains mid-size JSON strings (roughly 30K characters on average)

As of right now, the table has 53 users in it. About 20% of those have a negligible number of events, but the rest of the users have a fairly even smattering.

 

EXPLAIN (ANALYZE, BUFFERS) says:

                                                                          QUERY PLAN                                                                          

--------------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1218.57 rows=4000 width=615) (actual time=3352.390..3403.572 rows=4000 loops=1)  Buffers: shared hit=120244 read=160198

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..388147.29 rows=1274560 width=615) (actual time=3352.386..3383.100 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Filter: ((propogatorid <> '"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"'::text) AND (conflicted <> 1) AND (userid = '57dc984f1c87461c0967e228'::text))

         Rows Removed by Filter: 1685801

         Buffers: shared hit=120244 read=160198

 Planning time: 0.833 ms

 Execution time: 3407.633 ms

(9 rows)

If it matters/interests you, here is my underlying confusion:

From some internet sleuthing, I've decided that having a table per user (which would totally make this problem a non-issue) isn't a great idea. Because there is a file per table, having a table per user would not scale. My next thought was partial indexes (which would also totally help), but since there is also a table per index, this really doesn't side-step the problem. My rough mental model says: If there exists a way that a table-per-user scheme would make this more efficient, then there should also exist an index that could achieve the same effect (or close enough to not matter). I would think that "userid = '57dc984f1c87461c0967e228'" could utilize at least one of the two indexes on the userId column, but clearly I'm not understanding something.

Any help in making this query more efficient would be greatly appreciated, and any conceptual insights would be extra awesome.

Thanks for reading.

-Jake

----------------------

 

This stands out:  WHERE ID > 12468 AND propogatorId NOT IN ('"d8130ab9!-66d0!-4f13!-acec!-a9556362f0ad"')

As does this from the analyze:  Rows Removed by Filter: 1685801

 

The propogaterid is practically the only column NOT indexed and it’s used in a “not in”.  It looks like it’s having to do a table scan for all the rows above the id cutoff to see if any meet the filter requirement.  “not in” can be very expensive.  An index might help on this column.  Have you tried that?

 

Your rowcounts aren’t high enough to require partitioning or any other changes to your table that I can see right now.

 

Mike Sofen  (Synthetic Genomics)

 


Thanks Mike, that's true, I hadn't thought of non-indexed columns forcing a scan. Unfortunately, just to test this out, I tried pulling out the more suspect parts of the query, and it still seems to want to do an index scan:


EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERE userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;

                                                                        QUERY PLAN                                                                        

----------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..1140.62 rows=4000 width=615) (actual time=2706.365..2732.308 rows=4000 loops=1)

   Buffers: shared hit=120239 read=161924

   ->  Index Scan using syncerevent_pkey on syncerevent  (cost=0.43..364982.77 rows=1280431 width=615) (actual time=2706.360..2715.514 rows=4000 loops=1)

         Filter: (userid = '57dc984f1c87461c0967e228'::text)

         Rows Removed by Filter: 1698269

         Buffers: shared hit=120239 read=161924

 Planning time: 0.131 ms

 Execution time: 2748.526 ms

(8 rows)


It definitely looks to me like it's starting at the ID = 12468 row, and just grinding up the rows. The filter is (unsurprisingly) false for most of the rows, so it ends up having to chew up half the table before it actually finds 4000 rows that match.

After creating a partial index using that userId, things go way faster. This is more-or-less what I assumed I'd get by making having that multi-column index of (userId, Id), but alas:

remoteSyncerLogistics=> CREATE INDEX sillyIndex ON syncerevent (ID) where userId = '57dc984f1c87461c0967e228';

CREATE INDEX

remoteSyncerLogistics=> EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId = '57dc984f1c87461c0967e228' ORDER BY ID LIMIT 4000;

                                                                  QUERY PLAN                                                                  

----------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..443.21 rows=4000 width=615) (actual time=0.074..13.349 rows=4000 loops=1)

   Buffers: shared hit=842 read=13

   ->  Index Scan using sillyindex on syncerevent  (cost=0.43..141748.41 rows=1280506 width=615) (actual time=0.071..5.372 rows=4000 loops=1)

         Buffers: shared hit=842 read=13

 Planning time: 0.245 ms

 Execution time: 25.404 ms

(6 rows)


remoteSyncerLogistics=> EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId = '57dc984f1c87461c0967e228' AND ID > 12468 ORDER BY ID LIMIT 4000;

                                                                  QUERY PLAN                                                                  

----------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.43..453.34 rows=4000 width=615) (actual time=0.023..13.244 rows=4000 loops=1)

   Buffers: shared hit=855

   ->  Index Scan using sillyindex on syncerevent  (cost=0.43..144420.43 rows=1275492 width=615) (actual time=0.020..5.392 rows=4000 loops=1)

         Index Cond: (id > 12468)

         Buffers: shared hit=855

 Planning time: 0.253 ms

 Execution time: 29.371 ms

(7 rows)


Any thoughts?

-Jake

Hmmm, here's another unexpected piece of information:


EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERE userid = '57dc984f1c87461c0967e228';

                                                        QUERY PLAN                                                         

---------------------------------------------------------------------------------------------------------------------------

 Seq Scan on syncerevent  (cost=0.00..311251.51 rows=1248302 width=618) (actual time=0.008..4970.507 rows=1259137 loops=1)

   Filter: (userid = '57dc984f1c87461c0967e228'::text)

   Rows Removed by Filter: 2032685

   Buffers: shared hit=108601 read=161523

 Planning time: 0.092 ms

 Execution time: 7662.845 ms

(6 rows)


Looks like even just using userid='blah' doesn't actually result in the index being used, despite the fact that there are indexes on the userId column:


    "syncerevent_pkey" PRIMARY KEY, btree (id)

    "syncereventidindex" UNIQUE, btree (eventid)

    "anothersyncereventidindex" btree (userid)

    "anothersyncereventidindexwithascending" btree (userid, id)

    "asdfasdgasdf" btree (userid, id DESC)

    "syncereventuseridhashindex" hash (userid)


-Jake


So... it seems that setting the userId to one that has less rows in the table results in the index actually being used...

EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId = '57d35db7353b0d627c0e592f' AND ID > 12468 ORDER BY ID LIMIT 4000;

                                                                              QUERY PLAN                                                                                

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------

 Limit  (cost=0.56..8574.30 rows=4000 width=618) (actual time=0.031..13.190 rows=4000 loops=1)

   Buffers: shared hit=867

   ->  Index Scan using anothersyncereventidindexwithascending on syncerevent  (cost=0.56..216680.62 rows=101090 width=618) (actual time=0.027..5.313 rows=4000 loops=1)

         Index Cond: ((userid = '57d35db7353b0d627c0e592f'::text) AND (id > 12468))

         Buffers: shared hit=867

 Planning time: 0.168 ms

 Execution time: 29.331 ms

(7 rows)


Is there some way to force the use of one of the indexes on the userId column?

Re: Unexpected expensive index scan

От
Tom Lane
Дата:
[ Please don't re-quote the entire damn thread in each followup. Have
some respect for your readers' time, and assume that they have already
seen the previous traffic, or could go look it up if they haven't.
The point of quoting at all is just to quickly remind people where we
are in the discussion. ]

Jake Nielsen <jake.k.nielsen@gmail.com> writes:
> So... it seems that setting the userId to one that has less rows in the
> table results in the index actually being used...
> EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM SyncerEvent WHERe userId =
> '57d35db7353b0d627c0e592f' AND ID > 12468 ORDER BY ID LIMIT 4000;

It looks from the numbers floating around in this thread that the
userId used in your original query actually matches about 50% of
the table.  That would make it unsurprising that the planner doesn't
want to use an index.  A rule of thumb is that a seqscan is going
to be cheaper than an indexscan if your query retrieves, or even
just has to fetch, more than a few percent of the table.

Now, given the existence of an index on (userID, ID) --- in that
order --- I would expect the planner to want to use that index
for a query shaped exactly as you show above.  Basically, it knows
that that just requires starting at the ('57d35db7353b0d627c0e592f',
12468) position in the index and scanning forward for 4000 index
entries; no extraneous table rows will be fetched at all.  If you
increased the LIMIT enough, it'd go over to a seqscan-and-sort to
avoid doing so much random access to the table, but I'd think the
crossover point for that is well above 4000 out of 3.3M rows.

However, as soon as you add any other unindexable conditions,
the situation changes because rows that fail the additional
conditions represent useless fetches.  Now, instead of fetching
4000 rows using the index, it's fetching 4000 times some multiplier.

It's hard to tell for sure given the available info, but I think
that the extra inequalities in your original query reject a pretty
sizable proportion of rows, resulting in the indexscan approach
needing to fetch a great deal more than 4000 rows, making it look
to be more expensive than a seqscan.

I'm not sure why it's preferring the pkey index to the one on
(userID, ID), but possibly that has something to do with that
index being better correlated to the physical table order, resulting
in a prediction of less random I/O when using that index.

So the bottom line is that given your data statistics, there may
well be no really good plan for your original query.  It just
requires fetching a lot of rows, and indexes can't help very much.

If you say "well yeah, but it seems to perform fine when I force
it to use that index anyway", the answer may be that you need to
adjust random_page_cost.  The default value is OK for tables that
are mostly sitting on spinning rust, but if your database is
RAM-resident or SSD-resident you probably want a value closer to 1.

            regards, tom lane

Re: Unexpected expensive index scan

От
Jake Nielsen
Дата:


On Wed, Sep 28, 2016 at 6:04 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
[ Please don't re-quote the entire damn thread in each followup. Have
some respect for your readers' time, and assume that they have already
seen the previous traffic, or could go look it up if they haven't.
The point of quoting at all is just to quickly remind people where we
are in the discussion. ]

Sorry, understood.
 

If you say "well yeah, but it seems to perform fine when I force
it to use that index anyway", the answer may be that you need to
adjust random_page_cost.  The default value is OK for tables that
are mostly sitting on spinning rust, but if your database is
RAM-resident or SSD-resident you probably want a value closer to 1.

Ahhh, this could absolutely be the key right here. I could totally see why it would make sense for the planner to do what it's doing given that it's weighting sequential access more favorably than random access.

Beautiful! After changing the random_page_cost to 1.0 the original query went from ~3.5s to ~35ms. This is exactly the kind of insight I was fishing for in the original post. I'll keep in mind that the query planner is very tunable and has these sorts of hardware-related trade-offs in the future. I can't thank you enough!

Cheers!
 

Re: Unexpected expensive index scan

От
Jim Nasby
Дата:
On 9/28/16 1:11 PM, Jake Nielsen wrote:
> Beautiful! After changing the random_page_cost to 1.0 the original query
> went from ~3.5s to ~35ms. This is exactly the kind of insight I was
> fishing for in the original post. I'll keep in mind that the query
> planner is very tunable and has these sorts of hardware-related
> trade-offs in the future. I can't thank you enough!

Be careful with setting random_page_cost to exactly 1... that tells the
planner that an index scan has nearly the same cost as a sequential
scan, which is absolutely never the case, even with the database in
memory. 1.1 or maybe even 1.01 is probably a safer bet.

Also note that you can set those parameters within a single session, as
well as within a single transaction. So if you need to force a different
setting for a single query, you could always do

BEGIN;
SET LOCAL random_page_cost = 1;
SELECT ...
COMMIT; (or rollback...)
--
Jim Nasby, Data Architect, Blue Treble Consulting, Austin TX
Experts in Analytics, Data Architecture and PostgreSQL
Data in Trouble? Get it in Treble! http://BlueTreble.com
855-TREBLE2 (855-873-2532)   mobile: 512-569-9461