pg_stat_io clarifications: background worker, writes and reads

Hello list,

what is the "background worker" in the pg_stat_io statistics view? I'm
reading the documentation but can't figure this one out knowing that it is
not autovacuum or bgwriter. And I'm not aware of any extension I might
have with registered background worker.

Additionally, how can it be evictions > writes? I would expect every
eviction to cause a write.

Finally about "hits", I understand they are reads found in shared_buffers,
so they never registered into the "reads" counter. So is "reads" in
pg_stat_io the equivalent to misses, i.e. the opposite of "hits", the read
attempts not found in the shared_buffers, that needed to be fetched from
the disk (or OS buffercache)?

backend_type | object | context | reads | read_time |
writes | write_time | writebacks | writeback_time | extends | extend_time |
op_bytes | hits | evictions | reuses | fsyncs | fsync_time |
stats_reset

-------------------+---------------+---------+---------+-------------+--------+------------+------------+----------------+---------+-------------+----------+-----------+-----------+--------+--------+------------+-------------------------------
background worker | relation | normal | 5139575 | 2196288.011 |
63277 | 1766.94 | 0 | 0 | 0 | 0 |
8192 | 876913705 | 5139653 | | 0 | 0 | 2024-04-08
08:50:02.971192+00

Thank you in advance,
Dimitris

Hello list,

what is the "background worker" in the pg_stat_io statistics view? I'm
reading the documentation but can't figure this one out knowing that it is
not autovacuum or bgwriter. And I'm not aware of any extension I might
have with registered background worker.

Additionally, how can it be evictions > writes? I would expect every
eviction to cause a write.

Finally about "hits", I understand they are reads found in shared_buffers,
so they never registered into the "reads" counter. So is "reads" in
pg_stat_io the equivalent to misses, i.e. the opposite of "hits", the read
attempts not found in the shared_buffers, that needed to be fetched from
the disk (or OS buffercache)?

backend_type | object | context | reads | read_time |
writes | write_time | writebacks | writeback_time | extends | extend_time |
op_bytes | hits | evictions | reuses | fsyncs | fsync_time |
stats_reset

-------------------+---------------+---------+---------+-------------+--------+------------+------------+----------------+---------+-------------+----------+-----------+-----------+--------+--------+------------+-------------------------------
background worker | relation | normal | 5139575 | 2196288.011 |
63277 | 1766.94 | 0 | 0 | 0 | 0 |
8192 | 876913705 | 5139653 | | 0 | 0 | 2024-04-08
08:50:02.971192+00

Thank you in advance,
Dimitris

Hi,

In PostgreSQL, the pg_stat_io view provides detailed statistics on I/O operations. Background process perform maintenance tasks and other background operations essential to the functioning of the PostgreSQL database.
They include processes such as:

1. Autovacuum Workers
2. WAL Writer
3. Background Writer
4. Logical Replication Workers
5. Custom Background Workers

In the pg_stat_io view, statistics related to I/O operations performed by these background workers are recorded. 

Regards,
Imtiaz

On Wed, 15 May 2024, 01:26 Dimitrios Apostolou, <jimis@gmx.net> wrote:
Hello list,

what is the "background worker" in the pg_stat_io statistics view? I'm
reading the documentation but can't figure this one out knowing that it is
not autovacuum or bgwriter. And I'm not aware of any extension I might
have with registered background worker.

Additionally, how can it be evictions > writes? I would expect every
eviction to cause a write.

Finally about "hits", I understand they are reads found in shared_buffers,
so they never registered into the "reads" counter. So is "reads" in
pg_stat_io the equivalent to misses, i.e. the opposite of "hits", the read
attempts not found in the shared_buffers, that needed to be fetched from
the disk (or OS buffercache)?

    backend_type    |    object     | context |  reads  |  read_time  | writes | write_time | writebacks | writeback_time | extends | extend_time | op_bytes |   hits    | evictions | reuses | fsyncs | fsync_time |          stats_reset
-------------------+---------------+---------+---------+-------------+--------+------------+------------+----------------+---------+-------------+----------+-----------+-----------+--------+--------+------------+-------------------------------
  background worker | relation      | normal  | 5139575 | 2196288.011 |  63277 |    1766.94 |          0 |              0 |       0 |           0 |     8192 | 876913705 |   5139653 |        |      0 |          0 | 2024-04-08 08:50:02.971192+00

Thank you in advance,
Dimitris

So what is this particular "background worker" I'm seeing, given that I
have no replication or extensions?

Searching the logs I found entries like the following:

LOG: background worker "parallel worker" (PID 93384) exited with exit
code 1

This got logged when I killed a simple SELECT query that took too long
doing parallel seqscans. Could it be that the entry in pg_stat_io named
"background worker" also includes the parallel workers from a SELECT
query?

Thanks,
Dimitris

On Wed, 15 May 2024, Muhammad Imtiaz wrote:

Hi,

In PostgreSQL, the pg_stat_io view provides detailed statistics on I/O

operations. Background process perform maintenance tasks and other
background operations essential to the functioning of the PostgreSQL
database.

They include processes such as:

1. Autovacuum Workers
2. WAL Writer
3. Background Writer
4. Logical Replication Workers
5. Custom Background Workers

In the pg_stat_io view, statistics related to I/O operations performed

by these background workers are recorded.

Regards,
Imtiaz

On Wed, 15 May 2024, 01:26 Dimitrios Apostolou, <jimis@gmx.net> wrote:
Hello list,

what is the "background worker" in the pg_stat_io statistics view?

I'm

reading the documentation but can't figure this one out knowing

that it is

not autovacuum or bgwriter. And I'm not aware of any extension I

might

have with registered background worker.

Additionally, how can it be evictions > writes? I would expect

every

eviction to cause a write.

Finally about "hits", I understand they are reads found in

shared_buffers,

so they never registered into the "reads" counter. So is "reads" in
pg_stat_io the equivalent to misses, i.e. the opposite of "hits",

the read

attempts not found in the shared_buffers, that needed to be

fetched from

the disk (or OS buffercache)?

backend_type | object | context | reads |

read_time | writes | write_time | writebacks | writeback_time | extends |
extend_time | op_bytes | hits | evictions | reuses | fsyncs |
fsync_time | stats_reset

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

background worker | relation | normal | 5139575 |

2196288.011 | 63277 | 1766.94 | 0 | 0 | 0
| 0 | 8192 | 876913705 | 5139653 | | 0 |
0 | 2024-04-08 08:50:02.971192+00

Thank you in advance,
Dimitris

Hi
parallel worker are used for parallel execution of the queries and you can find the help in the below link.

https://www.postgresql.org/docs/current/how-parallel-query-works.html

Its controlled by following parameters.

max_worker_processes = 6
max_parallel_workers_per_gather = 3
max_parallel_workers = 6
The limit of concurrent parallel workers for the whole cluster is max_parallel_workers, which must be ≤ max_worker_processes. The limit of parallel workers per query is max_parallel_workers_per_gather.

Thanks
Kashif Zeeshan
Bitnine Global

On Wed, May 15, 2024 at 5:59 PM Dimitrios Apostolou <jimis@gmx.net> wrote:
So what is this particular "background worker" I'm seeing, given that I
have no replication or extensions?

Searching the logs I found entries like the following:

LOG:  background worker "parallel worker" (PID 93384) exited with exit code 1

This got logged when I killed a simple SELECT query that took too long
doing parallel seqscans. Could it be that the entry in pg_stat_io named
"background worker" also includes the parallel workers from a SELECT
query?

Thanks,
Dimitris