This repository has been archived by the owner on Feb 26, 2020. It is now read-only.
Write throttle wip #286
Closed
Write throttle wip #286
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
These kstat interfaces are required to port "Illumos #3537 want pool io kstats" to ZFS on Linux. kstat_waitq_enter() kstat_waitq_exit() kstat_runq_enter() kstat_runq_exit() Additionally, zero out the ks_data buffer in __kstat_create() so that the kstat_io_t counters are initialized to zero.
While porting Illumos #3537 I found that ks_lock member of kstat_t structure is different between Illumos and SPL. It is a pointer to the kmutex_t in Illumos, but the mutex lock itself in SPL. Apparently Illumos kstat API allows consumer to override the lock if required. With SPL implementation it is not possible anymore. Things were alright until the first attempt to actually override the lock. Porting of Illumos #3537 introduced such code for the first time. In order to provide the Solaris/Illumos like functionality we: 1. convert ks_lock to "kmutex_t *ks_lock" 2. create a new field "kmutex_t ks_private_lock" 3. On kstat_create() ks_lock = &ks_private_lock Thus if consumer doesn't care we still have our internal lock in use. If, however, consumer does care she has a chance to set ks_lock to anything else before calling kstat_install(). The rest of the code will use ks_lock regardless of its origin.
Needed for Illumos #3852. This interface is supposed to support a
variable-resolution timeout with nanosecond granularity. This
implementation just rounds up to jiffie resolution, as this was the most
expedient solution, and nanosecond resolution is rarely needed for
real-world performance tuning. Add flags from sys/callo.h as these are
used to control the behavior of cv_timedwait_hires(). Specifically,
CALLOUT_FLAG_ABSOLUTE
Normally, the expiration passed to the timeout API functions is
an expiration interval. If this flag is specified, then it is
interpreted as the expiration time itself.
CALLOUT_FLAG_ROUNDUP
Roundup the expiration time to the next resolution boundary. If this
flag is not specified, the expiration time is rounded down.
References:
https://www.illumos.org/issues/3582
illumos/illumos-gate@0689f76
This is needed for the Illumos #4045 write throttle patch. It is used in the arc eviction code to avoid blocking all arc activity by sitting on arcs_mtx too long.
behlendorf
pushed a commit
to behlendorf/spl
that referenced
this pull request
Oct 9, 2013
While porting Illumos #3537 I found that ks_lock member of kstat_t structure is different between Illumos and SPL. It is a pointer to the kmutex_t in Illumos, but the mutex lock itself in SPL. Apparently Illumos kstat API allows consumer to override the lock if required. With SPL implementation it is not possible anymore. Things were alright until the first attempt to actually override the lock. Porting of Illumos #3537 introduced such code for the first time. In order to provide the Solaris/Illumos like functionality we: 1. convert ks_lock to "kmutex_t *ks_lock" 2. create a new field "kmutex_t ks_private_lock" 3. On kstat_create() ks_lock = &ks_private_lock Thus if consumer doesn't care we still have our internal lock in use. If, however, consumer does care she has a chance to set ks_lock to anything else before calling kstat_install(). The rest of the code will use ks_lock regardless of its origin. Signed-off-by: Brian Behlendorf <[email protected]> Issue openzfs#286
behlendorf
pushed a commit
that referenced
this pull request
Oct 10, 2013
This is needed for the Illumos #4045 write throttle patch. It is used in the arc eviction code to avoid blocking all arc activity by sitting on arcs_mtx too long. Signed-off-by: Brian Behlendorf <[email protected]> Issue #286
|
The kpreempt() change has been merged in to master, and refreshed versions of the kstats changed have been added to #301. @nedbass when you refresh this with the reworked cv_timedwait_hires() patch can you open it as a new pull request and close this. I'm hopeful the kstat changes will be merged in the next week or so. |
behlendorf
pushed a commit
to behlendorf/spl
that referenced
this pull request
Oct 18, 2013
While porting Illumos #3537 I found that ks_lock member of kstat_t structure is different between Illumos and SPL. It is a pointer to the kmutex_t in Illumos, but the mutex lock itself in SPL. Apparently Illumos kstat API allows consumer to override the lock if required. With SPL implementation it is not possible anymore. Things were alright until the first attempt to actually override the lock. Porting of Illumos #3537 introduced such code for the first time. In order to provide the Solaris/Illumos like functionality we: 1. convert ks_lock to "kmutex_t *ks_lock" 2. create a new field "kmutex_t ks_private_lock" 3. On kstat_create() ks_lock = &ks_private_lock Thus if consumer doesn't care we still have our internal lock in use. If, however, consumer does care she has a chance to set ks_lock to anything else before calling kstat_install(). The rest of the code will use ks_lock regardless of its origin. Signed-off-by: Brian Behlendorf <[email protected]> Issue openzfs#286
behlendorf
pushed a commit
that referenced
this pull request
Oct 25, 2013
While porting Illumos #3537 I found that ks_lock member of kstat_t structure is different between Illumos and SPL. It is a pointer to the kmutex_t in Illumos, but the mutex lock itself in SPL. Apparently Illumos kstat API allows consumer to override the lock if required. With SPL implementation it is not possible anymore. Things were alright until the first attempt to actually override the lock. Porting of Illumos #3537 introduced such code for the first time. In order to provide the Solaris/Illumos like functionality we: 1. convert ks_lock to "kmutex_t *ks_lock" 2. create a new field "kmutex_t ks_private_lock" 3. On kstat_create() ks_lock = &ks_private_lock Thus if consumer doesn't care we still have our internal lock in use. If, however, consumer does care she has a chance to set ks_lock to anything else before calling kstat_install(). The rest of the code will use ks_lock regardless of its origin. Signed-off-by: Ned Bass <[email protected]> Signed-off-by: Brian Behlendorf <[email protected]> Issue #286
|
@nedbass When you get a chance can you rebase this branch against master with your final version of the cv_timedwait_hires() patch. There other three changes have now been merged in to master. |
|
Final versions of all the above patches have been merged. Closing. |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
SPL infrastructure needed for port of Illumos patch "4045 zfs write throttle & i/o scheduler performance work" and its dependencies.