More adaptive ARC eviction. #14359
Merged
More adaptive ARC eviction. #14359
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
amotin
added
Status: Code Review Needed
amotin
force-pushed
the
arc_evict
branch
2 times, most recently
from
0feb27f to
310fbc7
Compare
amotin
force-pushed
the
arc_evict
branch
3 times, most recently
from
8e24992 to
3b9dd0d
Compare
|
(sorry for the low-fidelity pic :) ) - I get a panic during import when I test this PR - looks like some incompatibility with L2ARC rebuild - |
@adamdmoss Thank you for the report. Appears I unexpectedly changed persistent L2ARC on-disk format. Added simple shim to fix it. |
|
Verified fixed - thanks! |
|
@amotin , thanks for making this. i currently try to test this. i have a question manpage is telling: what does a value of "500" exactly mean ? what should is set to maximise arc being used for metadata and avoid metadata eviction? proportional relation to what? |
@devZer0 It means data ghost hits cause 5 times smaller metadata cache reduction than metadata ghost hit cause data cache reduction. There is no upper limit. The higher you set it, the smaller pressure will be on metadata. It is not absolute, some metadata will likely be evicted, otherwise there will be no ghost state to indicate pressure, but after some time it should settle at some balance point where data and metadata ghost hits (read "almost cache hit but no") balance each other according to this coefficient. That is the whole point of being adaptive. |
allanjude
approved these changes
Feb 4, 2023
behlendorf
reviewed
Mar 2, 2023
amotin
force-pushed
the
arc_evict
branch
2 times, most recently
from
df3c3af to
2d948d6
Compare
|
While there I decided to remove unusable spa argument from arc_evict_impl() and reorder remaining more logically. |
behlendorf
approved these changes
Mar 2, 2023
behlendorf
added
Status: Accepted
Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Signed-off-by: Alexander Motin <[email protected]> Sponsored by: iXsystems, Inc.
|
I've noticed there is no more reason for arcstat_dnode_size to be an aggsum, since now it is read only once per arc_evict(), so I demoted it to cheaper wmsum. |
|
Merged. These changes worked as intended in my testing. |
12 tasks
mcmilk
pushed a commit
to mcmilk/zfs
that referenced
this pull request
Mar 13, 2023
Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Allan Jude <[email protected]> Signed-off-by: Alexander Motin <[email protected]> Sponsored by: iXsystems, Inc. Closes openzfs#14359
lundman
pushed a commit
to openzfsonwindows/openzfs
that referenced
this pull request
Mar 16, 2023
Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Allan Jude <[email protected]> Signed-off-by: Alexander Motin <[email protected]> Sponsored by: iXsystems, Inc. Closes openzfs#14359
|
Are these changes expected to be ported to 2.1.10? |
No. Same as few other ARC refactoring PRs of mine it will stay in 2.2. Those are quite a big and invasive change for a minor release. |
pcd1193182
pushed a commit
to pcd1193182/zfs
that referenced
this pull request
Sep 26, 2023
Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO. This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and meta- data, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata. The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places. This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits. Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable. Reviewed-by: Brian Behlendorf <[email protected]> Reviewed-by: Allan Jude <[email protected]> Signed-off-by: Alexander Motin <[email protected]> Sponsored by: iXsystems, Inc. Closes openzfs#14359
behlendorf
added a commit
that referenced
this pull request
Oct 13, 2023
New Features - Block cloning (#13392) - Linux container support (#14070, #14097, #12263) - Scrub error log (#12812, #12355) - BLAKE3 checksums (#12918) - Corrective "zfs receive" - Vdev and zpool user properties Performance - Fully adaptive ARC (#14359) - SHA2 checksums (#13741) - Edon-R checksums (#13618) - Zstd early abort (#13244) - Prefetch improvements (#14603, #14516, #14402, #14243, #13452) - General optimization (#14121, #14123, #14039, #13680, #13613, #13606, #13576, #13553, #12789, #14925, #14948) Signed-off-by: Brian Behlendorf <[email protected]>
lundman
pushed a commit
to openzfsonwindows/openzfs
that referenced
this pull request
Dec 12, 2023
New features: - Fully adaptive ARC eviction (openzfs#14359) - Block cloning (openzfs#13392) - Scrub error log (openzfs#12812, openzfs#12355) - Linux container support (openzfs#14070, openzfs#14097, openzfs#12263) - BLAKE3 Checksums (openzfs#12918) - Corrective "zfs receive" (openzfs#9372) Signed-off-by: Brian Behlendorf <[email protected]>
|
Hi @amotin thanks again so much for your work on this one. My tests on 2.2 (2.2.6) show a much more stable ARC size and use compared to the problematic prune storms on 2.1.x reported here: #9966 (comment). I did notice that there is some residue in the code and docs. If I understand correctly, the module parameter Lines 110 to 113 in e0039c7 Also in |
amotin
removed
the
Status: Design Review Needed
|
@gertvdijk You are right, it should be removed. Would you like to create a PR, or would prefer me to? About the openzfs-docs I have no idea, never touched it. I guess it could benefit from PR also. |
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.
Traditionally ARC adaptation was limited to MRU/MFU distribution. But for years people with metadata-centric workload demanded mechanisms to also manage data/metadata distribution, that in original ZFS was just a FIFO. As result ZFS effectively got separate states for data and metadata, minimum and maximum metadata limits etc, but it all required manual tuning, was not adaptive and in its heart remained a bad FIFO.
This change removes most of existing eviction logic, rewriting it from scratch. This makes MRU/MFU adaptation individual for data and metadata, same as the distribution between data and metadata themselves. Since most of required states separation was already done, it only required to make arcs_size state field specific per data/metadata.
The adaptation logic is still based on previous concept of ghost hits, just now it balances ARC capacity between 4 states: MRU data, MRU metadata, MFU data and MFU metadata. To simplify arc_c changes instead of arc_p measured in bytes, this code uses 3 variable arc_meta, arc_pd and arc_pm, representing ARC balance between metadata and data, MRU and MFU for data, and MRU and MFU for metadata respectively as 32-bit fixed point fractions. Since we care about the math result only when need to evict, this moves all the logic from arc_adapt() to arc_evict(), that reduces per-block overhead, since per-block operations are limited to stats collection, now moved from arc_adapt() to arc_access() and using cheaper wmsums. This also allows to remove ugly ARC_HDR_DO_ADAPT flag from many places.
This change also removes number of metadata specific tunables, part of which were actually not functioning correctly, since not all metadata are equal and some (like L2ARC headers) are not really evictable. Instead it introduced single opaque knob zfs_arc_meta_balance, tuning ARC's reaction on ghost hits, allowing administrator give more or less preference to metadata without setting strict limits.
Some of old code parts like arc_evict_meta() are just removed, because since introduction of ABD ARC they really make no sense: only headers referenced by small number of buffers are not evictable, and they are really not evictable no matter what this code do. Instead just call arc_prune_async() if too much metadata appear not evictable.
How Has This Been Tested?
Manually simulating different access pattern I was able to observe expected arc_meta, arc_pd and arc_pm changes.
Types of changes
Checklist:
Signed-off-by.