Update skc_obj_alloc for spl kmem caches that are backed by Linux #9474

sdimitro · 2019-10-16T20:35:24Z

Currently, for certain sizes and classes of allocations we use
SPL caches that are backed by caches in the Linux Slab allocator
to reduce fragmentation and increase utilization of memory. The
way things are implemented for these caches as of now though is
that we don't keep any statistics of the allocations that we
make from these caches.

This patch enables the tracking of allocated objects in those
SPL caches by making the trade-off of grabbing the cache lock
at every object allocation and free to update the respective
counter.

Additionally, this patch makes those caches visible in the
/proc/spl/kmem/slab special file.

As a side note, enabling the specific counter for those caches
enables SDB to create a more user-friendly interface than
/proc/spl/kmem/slab that can also cross-reference data from
slabinfo. Here is for example the output of one of those
caches in SDB that outputs the name of the underlying Linux
cache, the memory of SPL objects allocated in that cache,
and the percentage of those objects compared to all the
objects in it:

> spl_kmem_caches | filter obj.skc_name == "zio_buf_512" | pp
name        entry_size active_objs active_memory            source total_memory util
----------- ---------- ----------- ------------- ----------------- ------------ ----
zio_buf_512        512        2974         1.5MB kmalloc-512[SLUB]       16.9MB    8

Signed-off-by: Serapheim Dimitropoulos serapheim@delphix.com

How Has This Been Tested?

Besides running the test-suite to ensure no regression, I did some manual testing by double checking the values in the /proc filesystem and with sdb. I'm open to feedback on how to test this further.

Types of changes

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Performance enhancement (non-breaking change which improves efficiency)
Code cleanup (non-breaking change which makes code smaller or more readable)
Breaking change (fix or feature that would cause existing functionality to change)
Documentation (a change to man pages or other documentation)

Checklist:

My code follows the ZFS on Linux code style requirements.
I have updated the documentation accordingly.
I have read the contributing document.
I have added tests to cover my changes.
All new and existing tests passed.
All commit messages are properly formatted and contain Signed-off-by.

module/os/linux/spl/spl-kmem-cache.c

codecov · 2019-10-17T05:20:11Z

Codecov Report

Merging #9474 into master will decrease coverage by 0.28%.
The diff coverage is 33.33%.

@@            Coverage Diff             @@
##           master    #9474      +/-   ##
==========================================
- Coverage   79.28%      79%   -0.29%     
==========================================
  Files         415      415              
  Lines      123632   123640       +8     
==========================================
- Hits        98019    97678     -341     
- Misses      25613    25962     +349

Flag	Coverage Δ
#kernel	`79.7% <33.33%> (-0.09%)`	⬇️
#user	`66.49% <ø> (-0.82%)`	⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4313a5b...9a170fc. Read the comment docs.

Currently, for certain sizes and classes of allocations we use SPL caches that are backed by caches in the Linux Slab allocator to reduce fragmentation and increase utilization of memory. The way things are implemented for these caches as of now though is that we don't keep any statistics of the allocations that we make from these caches. This patch enables the tracking of allocated objects in those SPL caches by making the trade-off of grabbing the cache lock at every object allocation and free to update the respective counter. Additionally, this patch makes those caches visible in the /proc/spl/kmem/slab special file. As a side note, enabling the specific counter for those caches enables SDB to create a more user-friendly interface than /proc/spl/kmem/slab that can also cross-reference data from slabinfo. Here is for example the output of one of those caches in SDB that outputs the name of the underlying Linux cache, the memory of SPL objects allocated in that cache, and the percentage of those objects compared to all the objects in it: ``` > spl_kmem_caches | filter obj.skc_name == "zio_buf_512" | pp name ... source total_memory util ----------- ... ----------------- ------------ ---- zio_buf_512 ... kmalloc-512[SLUB] 16.9MB 8 ``` Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>

= Motivation `/proc/spl/kmem/slab` is a useful tool for debugging issues in ZoL that are memory-related. Unfortunately, it is not applicable to crash dumps and not very user friendly overall (e.g. you get almost no information for spl caches that are backed by Linux Slab caches). Besides the above, the specific use-case within Delphix is a replacement for `::kmastat` that we had in MDB in the illumos version of the product. The `spl_kmem_caches` command of this commit is implemented to complement the `slabs` command and cover the functionality of `::kmastat` that we miss further. = Patch This commit introduces the `spl` directory in our code base and implements `spl_kmem_caches`. By default the command iterates over all the SPL caches and prints the following statistics in human-readable form sorted by active_memory used: ``` name entry_size active_objs active_memory source total_memory util ------------------------ ---------- ----------- ------------- ------------------------------ ------------ ---- zio_data_buf_131072 135680 760 98.3MB zio_data_buf_131072[SPL ] 119.0MB 82 dnode_t 1168 25533 28.4MB dnode_t[SLUB] 28.6MB 99 zfs_znode_cache 1104 23039 24.3MB zfs_znode_cache[SLUB] 24.4MB 99 ... ``` All the properties can be seen in the help message and the user may specify the properties they are looking for with the `-o` option (and also sort them by a specific field using `-s`): ``` > spl_kmem_caches -h usage: spl_kmem_caches [-h] [-H] [-o FIELDS] [-p] [-r] [-s FIELD] [-v] optional arguments: -h, --help show this help message and exit -H do not display headers and separate fields by a single tab (scripted mode) -o FIELDS comma-separated list of fields to display -p display numbers in parseable (exact) values -r, --recursive recurse down children caches and print statistics -s FIELD sort rows by FIELD -v Print all statistics FIELDS := address, name, flags, object_size, entry_size, slab_size, objects_per_slab, entries_per_slab, slabs, active_slabs, active_memory, total_memory, objs, active_objs, inactive_objs, source, util If -o is not specified the default fields used are name, entry_size, active_objs, active_memory, source, total_memory, util. If the -s option is not specified and the command's output is not piped anywhere then we sort by the following fields in order: active_memory, address, name. If none of those exists in the field-set we sort by the first field specified in the set. > spl_kmem_caches -o "name,flags,slabs" -s slabs name flags slabs ------------------------ ----------------------- ----- zio_data_buf_131072 KMC_NODEBUG|KMC_VMEM 115 zio_buf_131072 KMC_NODEBUG|KMC_VMEM 25 zio_data_buf_40960 KMC_NODEBUG|KMC_VMEM 19 ... ``` Besides a PrettyPrinter, the command is also a Locator which means that the user can do things like this: ``` > spl_kmem_caches -s total_memory | head 1 | pp name entry_size active_objs active_memory source total_memory util ------------------- ---------- ----------- ------------- ------------------------- ------------ ---- zio_data_buf_131072 135680 760 98.3MB zio_data_buf_131072[SPL ] 119.0MB 82 ``` A couple more options mimicking Illumos conventions are the `-p` and `-H` options that output numbers to raw form and skip printing the headers respectively. This is generally useful for scripts or output that is to be processed later: ``` > spl_kmem_caches -H -p zio_data_buf_131072 135680 760 103116800 zio_data_buf_131072[SPL ] 124825600 82 dnode_t 1168 25573 29869264 dnode_t[SLUB] 30005920 99 zfs_znode_cache 1104 23039 25435056 zfs_znode_cache[SLUB] 25548768 99 ``` Note that for these command to fully work for SPL caches that are backed by Linux Slab caches we need the appropriate ZoL changes that enables us to track allocations for those cases (see PR: openzfs/zfs#9474). If the commit from that PR is not in the running system then all of these caches will incorectly be reported empty. Side-change: As the `spl_kmem_caches` command is very close both in terms of implementation and functionality with the `slabs` command this patch also updates some of the columns and calculations in the output of `slabs`, to make the commands consistent with each other. = Testing/Verification I verified the values of all their fields in the following ways: [1] For the fields that are also visible in /proc/slabinfo, I ensured that the values matched. [2] For anything else I did the math and cross-referenced all the values ensuring that they are consistent with each other. Besides the manual testing done shown in the output above I also ensured the following error cases: * ask for invalid field name ``` > spl_kmem_caches -o bogus sdb: spl_kmem_caches: 'bogus' is not a valid field ``` * sort by invalid field name ``` sdb: spl_kmem_caches: invalid input: 'bogus' is not in field set (name, entry_size, active_objs, active_memory, source, total_memory, util) ``` * attempt to sort by field that is not in the requested field set ``` > spl_kmem_caches -o "name,address" -s objs sdb: spl_kmem_caches: invalid input: 'objs' is not in field set (name, address) ``` = Other Notes The command only works with SLUB and until we have proper lexing we still pass fields like this: `> spl_kmem_caches -o "name,address"`

Currently, for certain sizes and classes of allocations we use SPL caches that are backed by caches in the Linux Slab allocator to reduce fragmentation and increase utilization of memory. The way things are implemented for these caches as of now though is that we don't keep any statistics of the allocations that we make from these caches. This patch enables the tracking of allocated objects in those SPL caches by making the trade-off of grabbing the cache lock at every object allocation and free to update the respective counter. Additionally, this patch makes those caches visible in the /proc/spl/kmem/slab special file. As a side note, enabling the specific counter for those caches enables SDB to create a more user-friendly interface than /proc/spl/kmem/slab that can also cross-reference data from slabinfo. Here is for example the output of one of those caches in SDB that outputs the name of the underlying Linux cache, the memory of SPL objects allocated in that cache, and the percentage of those objects compared to all the objects in it: ``` > spl_kmem_caches | filter obj.skc_name == "zio_buf_512" | pp name ... source total_memory util ----------- ... ----------------- ------------ ---- zio_buf_512 ... kmalloc-512[SLUB] 16.9MB 8 ``` Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com> Closes openzfs#9474

Currently, for certain sizes and classes of allocations we use SPL caches that are backed by caches in the Linux Slab allocator to reduce fragmentation and increase utilization of memory. The way things are implemented for these caches as of now though is that we don't keep any statistics of the allocations that we make from these caches. This patch enables the tracking of allocated objects in those SPL caches by making the trade-off of grabbing the cache lock at every object allocation and free to update the respective counter. Additionally, this patch makes those caches visible in the /proc/spl/kmem/slab special file. As a side note, enabling the specific counter for those caches enables SDB to create a more user-friendly interface than /proc/spl/kmem/slab that can also cross-reference data from slabinfo. Here is for example the output of one of those caches in SDB that outputs the name of the underlying Linux cache, the memory of SPL objects allocated in that cache, and the percentage of those objects compared to all the objects in it: ``` > spl_kmem_caches | filter obj.skc_name == "zio_buf_512" | pp name ... source total_memory util ----------- ... ----------------- ------------ ---- zio_buf_512 ... kmalloc-512[SLUB] 16.9MB 8 ``` Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com> Closes #9474

Currently, for certain sizes and classes of allocations we use SPL caches that are backed by caches in the Linux Slab allocator to reduce fragmentation and increase utilization of memory. The way things are implemented for these caches as of now though is that we don't keep any statistics of the allocations that we make from these caches. This patch enables the tracking of allocated objects in those SPL caches by making the trade-off of grabbing the cache lock at every object allocation and free to update the respective counter. Additionally, this patch makes those caches visible in the /proc/spl/kmem/slab special file. As a side note, enabling the specific counter for those caches enables SDB to create a more user-friendly interface than /proc/spl/kmem/slab that can also cross-reference data from slabinfo. Here is for example the output of one of those caches in SDB that outputs the name of the underlying Linux cache, the memory of SPL objects allocated in that cache, and the percentage of those objects compared to all the objects in it: ``` > spl_kmem_caches | filter obj.skc_name == "zio_buf_512" | pp name ... source total_memory util ----------- ... ----------------- ------------ ---- zio_buf_512 ... kmalloc-512[SLUB] 16.9MB 8 ``` Reviewed-by: Matt Ahrens <matt@delphix.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com> Closes openzfs#9474

sdimitro force-pushed the zol-spl-kmem-counters branch 2 times, most recently from fbc475c to 3a7308f Compare October 16, 2019 20:50

sdimitro requested review from behlendorf and ahrens October 16, 2019 20:59

ahrens approved these changes Oct 16, 2019

View reviewed changes

sdimitro mentioned this pull request Oct 16, 2019

Introduce spl_kmem_caches command delphix/sdb#42

Merged

behlendorf added the Status: Code Review Needed Ready for review and testing label Oct 17, 2019

behlendorf requested changes Oct 17, 2019

View reviewed changes

module/os/linux/spl/spl-kmem-cache.c Outdated Show resolved Hide resolved

sdimitro force-pushed the zol-spl-kmem-counters branch from 3a7308f to dd9049a Compare October 17, 2019 16:42

sdimitro force-pushed the zol-spl-kmem-counters branch from dd9049a to 9a170fc Compare October 17, 2019 17:01

behlendorf approved these changes Oct 18, 2019

View reviewed changes

behlendorf added Status: Accepted Ready to integrate (reviewed, tested) and removed Status: Code Review Needed Ready for review and testing labels Oct 18, 2019

behlendorf merged commit 851eda3 into openzfs:master Oct 18, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update skc_obj_alloc for spl kmem caches that are backed by Linux #9474

Update skc_obj_alloc for spl kmem caches that are backed by Linux #9474

sdimitro commented Oct 16, 2019

codecov bot commented Oct 17, 2019 •

edited

Loading

Update skc_obj_alloc for spl kmem caches that are backed by Linux #9474

Update skc_obj_alloc for spl kmem caches that are backed by Linux #9474

Conversation

sdimitro commented Oct 16, 2019

How Has This Been Tested?

Types of changes

Checklist:

codecov bot commented Oct 17, 2019 • edited Loading

Codecov Report

codecov bot commented Oct 17, 2019 •

edited

Loading