Wiki editor request - Mailing list pgsql-www
| From | yigong hu |
|---|---|
| Subject | Wiki editor request |
| Date | |
| Msg-id | CAOxFffdaBFkCryxhutHAtq9WTxiuxjpVW8Utuq0=4rCPhuzpig@mail.gmail.com Whole thread Raw |
| Responses |
Re: Wiki editor request
|
| List | pgsql-www |
Hi,
Would it be possible to get editor permission on the Postgres wiki? My community account name is yigonghu.
Here is a minor change I’ve wanted to make as I browse the wiki:
The PostgreSQL manual contains the following paragraph https://wiki.postgresql.org/wiki/Tuning_Your_PostgreSQL_Server about the log_statement configuration:
> "Using all in production leads to severe performance penalties."
The description mentions that "all" log_statement would cause performance penalties in production mode but in the testing, I found that setting log_statement to all will not cause performance loss under write-intensive workload unless the synchronous_commit is off. Also, it seems that setting log_statement to another value wouldn' cause performance regression but in my testing, I found that the "mod" log_statement would cause a performance loss if the synchronous_commit is off. Thus, I would suggest to elaborate on this description further as:
> "Using all or mod in production would lead to severe performance penalties under some workload. Turning the synchronous_commit to off would cause a more severe performance regression"
Rationale:
I test the log_statement on PostgreSQL 11.0. The PostgreSQL is running on hardware: 40 vCPU, 64 GB ram, 10GB network, and 1TB HDD. The OS is ubuntu 16.04.
I first run a read-intensive workload with different log_statment setting. The result shows that for all log_statment, the average latency per query is 7.18ms. For the "mod" statement, the average latency per query is 2.9ms. For the "none" log_statement, the average latency per query is 2.95ms. The result is consistent with the description that "all" log statement would cause performance penalties. However, when I test on a write-intensive workload, the result shows that for the "all" log_statment, the average latency per query is 8.51ms. For the "mod" statement, the average latency per query is 8.53ms. For the "none" log_statement, the average latency per query is 8.68ms. It seems that there are no performance penalties for the write workload if you are using "all" log statement.
Also, if I set the synchronous_commit to off and rerun the write-intensive workload, For the "all" log_statment, the average latency per query is 0.71ms. For the "mod" statement, the average latency per query is 0.71ms. For the "none" log_statement, the average latency per query is 0.19ms. This seems to suggest that if closing the synchronous_commit would introduce performance penalties to all and mod log statements.
Would it be possible to get editor permission on the Postgres wiki? My community account name is yigonghu.
Here is a minor change I’ve wanted to make as I browse the wiki:
The PostgreSQL manual contains the following paragraph https://wiki.postgresql.org/wiki/Tuning_Your_PostgreSQL_Server about the log_statement configuration:
> "Using all in production leads to severe performance penalties."
The description mentions that "all" log_statement would cause performance penalties in production mode but in the testing, I found that setting log_statement to all will not cause performance loss under write-intensive workload unless the synchronous_commit is off. Also, it seems that setting log_statement to another value wouldn' cause performance regression but in my testing, I found that the "mod" log_statement would cause a performance loss if the synchronous_commit is off. Thus, I would suggest to elaborate on this description further as:
> "Using all or mod in production would lead to severe performance penalties under some workload. Turning the synchronous_commit to off would cause a more severe performance regression"
Rationale:
I test the log_statement on PostgreSQL 11.0. The PostgreSQL is running on hardware: 40 vCPU, 64 GB ram, 10GB network, and 1TB HDD. The OS is ubuntu 16.04.
I first run a read-intensive workload with different log_statment setting. The result shows that for all log_statment, the average latency per query is 7.18ms. For the "mod" statement, the average latency per query is 2.9ms. For the "none" log_statement, the average latency per query is 2.95ms. The result is consistent with the description that "all" log statement would cause performance penalties. However, when I test on a write-intensive workload, the result shows that for the "all" log_statment, the average latency per query is 8.51ms. For the "mod" statement, the average latency per query is 8.53ms. For the "none" log_statement, the average latency per query is 8.68ms. It seems that there are no performance penalties for the write workload if you are using "all" log statement.
Also, if I set the synchronous_commit to off and rerun the write-intensive workload, For the "all" log_statment, the average latency per query is 0.71ms. For the "mod" statement, the average latency per query is 0.71ms. For the "none" log_statement, the average latency per query is 0.19ms. This seems to suggest that if closing the synchronous_commit would introduce performance penalties to all and mod log statements.