reactor, linux-aio: advise users in more detail on setting aio-max-nr
#2396
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
|
Some example outputs, through scylla (all wrapped here manually, for readability):
|
|
i will take a look at this PR tomorrow. |
tchaikov
reviewed
Aug 26, 2024
src/core/reactor.cc
Outdated
| seastar_logger.info(""); | ||
| log_aiocbs(log_level::info, storage_iocbs, preempt_iocbs, network_iocbs); | ||
| seastar_logger.info(""); | ||
| seastar_logger.info("Available AIO control blocks = aio-max-nr - aio-nr = {} - {} = {}", aio_max_nr, aio_nr, available_aio); |
There was a problem hiding this comment.
the default logging level is set to info, which ensures this message is always printed. i am wondering if we need this logging message if the number of aio cb is large enough. because i think, "no news is good news", which potentially reduces unnecessary verbosity.
so, maybe we can print it only when the number of aio cb is not sufficient and might hurt the network effienciency of the seastar application or it is not large enough to fulfill the minimum requirement?
There was a problem hiding this comment.
This patch (for seastar code) is also supposed to replace the following passage in scylladb's docs/dev/docker-hub.md (as of scylladb commit ff52527c54e8):
-Using multiple cores requires setting a proper value to the `/proc/sys/fs/aio-max-nr`.
-On many non-production systems, it will be equal to 65K. The formula
-to calculate the proper value is:
-
- Available AIO on the system - (request AIO per-cpu * ncpus) =
- aio_max_nr - aio_nr < (reactor::max_aio + detect_aio_poll + reactor_backend_aio::max_polls) * cpu_cores =
- aio_max_nr - aio_nr < (1024 + 2 + 10000) * cpu_cores =
- aio_max_nr - aio_nr < 11026 * cpu_cores
-
- where
-
- reactor::max_aio = max_aio_per_queue * max_queues,
- max_aio_per_queue = 128,
- max_queues = 8.
-
(See here why -- it's just too stale, and seeing actual numbers is better than purely symbolic formulae.)
Given that the static documentation is going away, I believe that the basic info should be printed even if the resources prove sufficint. However, I do agree that seeing this message at every startup, with default log level configuration, becomes boring pretty fast. I suggest: downgrade the log level from info to verbose. Would that work for you?
There was a problem hiding this comment.
@lersek that'd work. thanks!
There was a problem hiding this comment.
correction: the log level just below info is debug, not verbose (I remembered verbose from another project, sorry). ... that is, I'm going to change the log level to debug.
Print a table of the AIO counts we intend to use (at "debug" level). If the intended counts cannot be allocated, reduce the network AIO count like before, but now also print an updated table (as a warning). Tell the user the specific number they should write to "/proc/sys/fs/aio-max-nr" for avoiding the reduction (assuming their current "aio-nr" does not change). If the per CPU network AIO count would have to be reduced under 1, format a clear error message with (up to) four specific recommendations: - aio-max-nr value for reaching 1 network AIOCB per CPU (always printed), - aio-max-nr value for reaching the originally requested network AIOCB per CPU (always printed), - the highest value to reduce the application's CPU count to, in order to reach 1 network AIOCB per CPU, within the current (aio-max-nr - aio-nr) budget (printed only if this SMP count is at least 1), - the value to reduce the application's CPU count to, in order to reach the originally requested network AIOCB per CPU, within the current (aio-max-nr - aio-nr) budget (printed only if this SMP count is at least 1). Motivated by <scylladb/scylladb#5981>. Signed-off-by: Laszlo Ersek <[email protected]>
|
v2:
example output (pass |
lersek
added a commit
to lersek/scylla
that referenced
this pull request
Sep 2, 2024
What we have today in "docs/dev/docker-hub.md" on "aio-max-nr" dates back to scylla commit f441202 ("docs/docker-hub.md: add quickstart section with --smp 1", 2020-09-22). Problems with the current language: - The "65K" claim as default value on non-production systems is wrong; "fs/aio.c" in Linux initializes "aio_max_nr" to 0x10000, which is 64K. - The section in question uses equal signs (=) incorrectly. The intent was probably to say "which means the same as", but that's not what equality means. - In the same section, the relational operator "<" is bogus. The available AIO count must be at least as high (>=) as the requested AIO count. - Clearer names should be used; adjust_max_networking_aio_io_control_blocks() in "src/core/reactor.cc" sets a great example: - "reactor::max_aio" should be called "storage_iocbs"; - "detect_aio_poll" should be called "preempt_iocbs"; - "reactor_backend_aio::max_polls" should be called "network_iocbs"; - The specific value 10000 for the last one ("network_iocbs") is not correct in scylla's context. It is correct as the Seastar default, but scylla has used 50000 since commit 2cfc517 ("main, test: adjust number of networking iocbs", 2021-07-18). Rewrite the section to address these problems. See also: - scylladb#5981 - scylladb/seastar#2396 - scylladb#19921 Signed-off-by: Laszlo Ersek <[email protected]>
lersek
added a commit
to lersek/scylla
that referenced
this pull request
Sep 2, 2024
What we have today in "docs/dev/docker-hub.md" on "aio-max-nr" dates back to scylla commit f441202 ("docs/docker-hub.md: add quickstart section with --smp 1", 2020-09-22). Problems with the current language: - The "65K" claim as default value on non-production systems is wrong; "fs/aio.c" in Linux initializes "aio_max_nr" to 0x10000, which is 64K. - The section in question uses equal signs (=) incorrectly. The intent was probably to say "which means the same as", but that's not what equality means. - In the same section, the relational operator "<" is bogus. The available AIO count must be at least as high (>=) as the requested AIO count. - Clearer names should be used; adjust_max_networking_aio_io_control_blocks() in "src/core/reactor.cc" sets a great example: - "reactor::max_aio" should be called "storage_iocbs", - "detect_aio_poll" should be called "preempt_iocbs", - "reactor_backend_aio::max_polls" should be called "network_iocbs". - The specific value 10000 for the last one ("network_iocbs") is not correct in scylla's context. It is correct as the Seastar default, but scylla has used 50000 since commit 2cfc517 ("main, test: adjust number of networking iocbs", 2021-07-18). Rewrite the section to address these problems. See also: - scylladb#5981 - scylladb/seastar#2396 - scylladb#19921 Signed-off-by: Laszlo Ersek <[email protected]>
avikivity
added a commit
to scylladb/scylladb
that referenced
this pull request
Sep 9, 2024
…aszlo Ersek ~~~ What we have today in "docs/dev/docker-hub.md" on "aio-max-nr" dates back to scylla commit f441202 ("docs/docker-hub.md: add quickstart section with --smp 1", 2020-09-22). Problems with the current language: - The "65K" claim as default value on non-production systems is wrong; "fs/aio.c" in Linux initializes "aio_max_nr" to 0x10000, which is 64K. - The section in question uses equal signs (=) incorrectly. The intent was probably to say "which means the same as", but that's not what equality means. - In the same section, the relational operator "<" is bogus. The available AIO count must be at least as high (>=) as the requested AIO count. - Clearer names should be used; adjust_max_networking_aio_io_control_blocks() in "src/core/reactor.cc" sets a great example: - "reactor::max_aio" should be called "storage_iocbs", - "detect_aio_poll" should be called "preempt_iocbs", - "reactor_backend_aio::max_polls" should be called "network_iocbs". - The specific value 10000 for the last one ("network_iocbs") is not correct in scylla's context. It is correct as the Seastar default, but scylla has used 50000 since commit 2cfc517 ("main, test: adjust number of networking iocbs", 2021-07-18). Rewrite the section to address these problems. See also: - #5981 - scylladb/seastar#2396 - #19921 Signed-off-by: Laszlo Ersek <[email protected]> ~~~ No need for backporting; the documentation being refreshed targets developers as audience, not end-users. Closes #20398 * github.com:scylladb/scylladb: docs/dev/docker-hub.md: refresh aio-max-nr calculation docs/dev/docker-hub.md: strip trailing whitespace
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.
Print a table of the AIO counts we intend to use (at "debug" level).
If the intended counts cannot be allocated, reduce the network AIO count like before, but now also print an updated table (as a warning). Tell the user the specific number they should write to
/proc/sys/fs/aio-max-nrfor avoiding the reduction (assuming their current "aio-nr" does not change).If the per CPU network AIO count would have to be reduced under 1, format a clear error message with (up to) four specific recommendations:
aio-max-nrvalue for reaching 1 network AIOCB per CPU (always printed),aio-max-nrvalue for reaching the originally requested network AIOCB per CPU (always printed),(aio-max-nr - aio-nr)budget (printed only if this SMP count is at least 1),(aio-max-nr - aio-nr)budget (printed only if this SMP count is at least 1).Motivated by scylladb/scylladb#5981.