Don't flush at each put_copy_data call, but flush at get_result #462
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This better mimics, what libpq does internally. put_copy_data is significantly faster when it doesn't flush at every call. This is by a factor of 4 on Linux and 10 on Windows when sending typical per-row blocks of 60 byte. put_copy_end unconditionally calls flush in libpq, so it is not changed here. Also adjust conn.block to send all enqueued data to mimic the behavior of get_result in libpq. With the change to put_copy_data, unsent data can happen, when largs blocks are sent. In this case get_result should catch up on flush. This is what the newly added spec verifies. PQgetResult does flushing based on it's internal states that we don't have access to. Since conn.block is performance critical in case of single_row_mode, we don't flush at every call to conn.block, but only when it's about to wait for IO.
larskanis
added a commit
to larskanis/ruby-pg
that referenced
this pull request
Aug 8, 2022
We had a blocking flush in pg-1.3.x at every call to put_copy_data. This made sure, that all data is sent until the next put_copy_data. In ged#462 (and pg-1.4.0 to .2) the behaviour was changed to rely on the non-blocking flushs libpq is doing internally. This makes a decent performance improvement especially on Windows. Unfortunately ged#473 proved that memory bloat can happen, when sending the data is slower than calls to put_copy_data happen. As a trade-off this proposes to do a blocking flush only every 100 calls. If libpq is running in blocking mode (PG::Connection.async_api = false) put_copy_data does a blocking flush every time new memory is allocated. Unfortunately we don't have this kind of information, since we don't have access to libpq's PGconn struct and the return codes don't give us an indication when this happens. So doing a flush at every fixed number of calls is a very simple heuristic. Fixes ged#473
This was referenced Aug 8, 2022
larskanis
added a commit
to larskanis/ruby-pg
that referenced
this pull request
Aug 9, 2022
We had a blocking flush in pg-1.3.x at every call to put_copy_data. This made sure, that all data is sent until the next put_copy_data. In ged#462 (and pg-1.4.0 to .2) the behaviour was changed to rely on the non-blocking flushs libpq is doing internally. This makes a decent performance improvement especially on Windows. Unfortunately ged#473 proved that memory bloat can happen, when sending the data is slower than calls to put_copy_data happen. As a trade-off this proposes to do a blocking flush only every 100 calls. If libpq is running in blocking mode (PG::Connection.async_api = false) put_copy_data does a blocking flush every time new memory is allocated. Unfortunately we don't have this kind of information, since we don't have access to libpq's PGconn struct and the return codes don't give us an indication when this happens. So doing a flush at every fixed number of calls is a very simple heuristic. Fixes ged#473
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This better mimics, what libpq does internally.
put_copy_datais significantly faster when it doesn'tflushat every call. This is by a factor of 4 on Linux and 10 on Windows when sending typical per-row blocks of 60 byte.put_copy_endunconditionally callsflushin libpq, so it is not changed here.Also adjust
conn.blockto send all enqueued data to mimic the behavior ofget_resultin libpq. With the change toput_copy_data, unsent data can happen, when largs blocks are sent. In this caseget_resultshould catch up on flush. This is what the newly added spec verifies.PQgetResultdoes flushing based on it's internal states that we don't have access to. Sinceconn.blockis performance critical in case ofsingle_row_mode, we don't flush at every call toconn.block, but only when it's about to wait for IO.