I retested the issue on lustre 2.15.4_4 with ZFS 2.2.3/4 and I can no longer reproduce the issue.
At this point we can close this issues and explain that starting at ZFS 2.2.3 the issue is fixed.

Adding the arc_summary for the oom:
arc_summary

------------------------------------------------------------------------
ZFS Subsystem Report Tue Feb 13 12:22:54 2024
Linux 4.18.0-513.11.1.1toss.t4.x86_64 2.1.14_1llnl-1
Machine: garter5 (x86_64) 2.1.14_1llnl-1

ARC status: HEALTHY
Memory throttle count: 0

ARC size (current): 123.8 % 115.9 GiB
Target size (adaptive): 100.0 % 93.6 GiB
Min size (hard limit): 6.2 % 5.9 GiB
Max size (high water): 16:1 93.6 GiB
Most Frequently Used (MFU) cache size: 66.7 % 6.0 GiB
Most Recently Used (MRU) cache size: 33.3 % 3.0 GiB
Metadata cache size (hard limit): 75.0 % 70.2 GiB
Metadata cache size (current): 1.1 % 794.2 MiB
Dnode cache size (hard limit): 10.0 % 7.0 GiB
Dnode cache size (current): 0.2 % 16.5 MiB

ARC hash breakdown:
Elements max: 394.5k
Elements current: 95.8 % 377.8k
Collisions: 10.1k
Chain max: 2
Chains: 2.2k

ARC misc:
Deleted: 731.6k
Mutex misses: 4.3M
Eviction skips: 325.9M
Eviction skips due to L2 writes: 0
L2 cached evictions: 0 Bytes
L2 eligible evictions: 200.2 GiB
L2 eligible MFU evictions: 72.7 % 145.5 GiB
L2 eligible MRU evictions: 27.3 % 54.7 GiB
L2 ineligible evictions: 716.4 GiB

ARC total accesses (hits + misses): 7.5M
Cache hit ratio: 81.0 % 6.0M
Cache miss ratio: 19.0 % 1.4M
Actual hit ratio (MFU + MRU hits): 80.9 % 6.0M
Data demand efficiency: 58.3 % 319.8k
Data prefetch efficiency: 0.4 % 1.0M

Cache hits by cache type:
Most frequently used (MFU): 87.3 % 5.3M
Most recently used (MRU): 12.7 % 763.4k
Most frequently used (MFU) ghost: 2.8 % 171.0k
Most recently used (MRU) ghost: 2.3 % 139.8k

Cache hits by data type:
Demand data: 3.1 % 186.3k
Prefetch data: 0.1 % 3.9k
Demand metadata: 96.8 % 5.8M
Prefetch metadata: < 0.1 % 469

Cache misses by data type:
Demand data: 9.4 % 133.4k
Prefetch data: 71.3 % 1.0M
Demand metadata: 18.9 % 268.9k
Prefetch metadata: 0.3 % 4.6k

DMU prefetch efficiency: 926.8k
Hit ratio: 86.2 % 799.0k
Miss ratio: 13.8 % 127.8k

L2ARC not detected, skipping section

Solaris Porting Layer (SPL):
spl_hostid 0
spl_hostid_path /etc/hostid
spl_kmem_alloc_max 1048576
spl_kmem_alloc_warn 65536
spl_kmem_cache_kmem_threads 4
spl_kmem_cache_magazine_size 0
spl_kmem_cache_max_size 32
spl_kmem_cache_obj_per_slab 8
spl_kmem_cache_reclaim 0
spl_kmem_cache_slab_limit 16384
spl_max_show_tasks 512
spl_panic_halt 0
spl_schedule_hrtimeout_slack_us 0
spl_taskq_kick 0
spl_taskq_thread_bind 0
spl_taskq_thread_dynamic 0
spl_taskq_thread_priority 1
spl_taskq_thread_sequential 4

Tunables:
dbuf_cache_hiwater_pct 10
dbuf_cache_lowater_pct 10
dbuf_cache_max_bytes 18446744073709551615
dbuf_cache_shift 5
dbuf_metadata_cache_max_bytes 18446744073709551615
dbuf_metadata_cache_shift 6
dbuf_mutex_cache_shift 0
dmu_object_alloc_chunk_shift 7
dmu_prefetch_max 134217728
ignore_hole_birth 1
l2arc_exclude_special 0
l2arc_feed_again 1
l2arc_feed_min_ms 200
l2arc_feed_secs 1
l2arc_headroom 2
l2arc_headroom_boost 200
l2arc_meta_percent 33
l2arc_mfuonly 0
l2arc_noprefetch 1
l2arc_norw 0
l2arc_rebuild_blocks_min_l2size 1073741824
l2arc_rebuild_enabled 1
l2arc_trim_ahead 0
l2arc_write_boost 8388608
l2arc_write_max 8388608
metaslab_aliquot 1048576
metaslab_bias_enabled 1
metaslab_debug_load 0
metaslab_debug_unload 0
metaslab_df_max_search 16777216
metaslab_df_use_largest_segment 0
metaslab_force_ganging 16777217
metaslab_fragmentation_factor_enabled 1
metaslab_lba_weighting_enabled 1
metaslab_preload_enabled 1
metaslab_unload_delay 32
metaslab_unload_delay_ms 600000
send_holes_without_birth_time 1
spa_asize_inflation 24
spa_config_path /etc/zfs/zpool.cache
spa_load_print_vdev_tree 0
spa_load_verify_data 1
spa_load_verify_metadata 1
spa_load_verify_shift 4
spa_slop_shift 5
vdev_file_logical_ashift 9
vdev_file_physical_ashift 9
vdev_removal_max_span 32768
vdev_validate_skip 0
zap_iterate_prefetch 1
zfetch_array_rd_sz 67108864
zfetch_max_distance 67108864
zfetch_max_idistance 67108864
zfetch_max_sec_reap 2
zfetch_max_streams 500
zfetch_min_distance 4194304
zfetch_min_sec_reap 1
zfs_abd_scatter_enabled 1
zfs_abd_scatter_max_order 10
zfs_abd_scatter_min_size 1536
zfs_admin_snapshot 0
zfs_allow_redacted_dataset_mount 0
zfs_arc_average_blocksize 8192
zfs_arc_dnode_limit 0
zfs_arc_dnode_limit_percent 10
zfs_arc_dnode_reduce_percent 10
zfs_arc_evict_batch_limit 10
zfs_arc_eviction_pct 200
zfs_arc_grow_retry 0
zfs_arc_lotsfree_percent 10
zfs_arc_max 0
zfs_arc_meta_adjust_restarts 4096
zfs_arc_meta_limit 0
zfs_arc_meta_limit_percent 75
zfs_arc_meta_min 0
zfs_arc_meta_prune 10000
zfs_arc_meta_strategy 1
zfs_arc_min 0
zfs_arc_min_prefetch_ms 0
zfs_arc_min_prescient_prefetch_ms 0
zfs_arc_p_dampener_disable 1
zfs_arc_p_min_shift 0
zfs_arc_pc_percent 0
zfs_arc_prune_task_threads 1
zfs_arc_shrink_shift 0
zfs_arc_shrinker_limit 10000
zfs_arc_sys_free 0
zfs_async_block_max_blocks 18446744073709551615
zfs_autoimport_disable 1
zfs_btree_verify_intensity 0
zfs_checksum_events_per_second 20
zfs_commit_timeout_pct 5
zfs_compressed_arc_enabled 1
zfs_condense_indirect_commit_entry_delay_ms 0
zfs_condense_indirect_obsolete_pct 25
zfs_condense_indirect_vdevs_enable 1
zfs_condense_max_obsolete_bytes 1073741824
zfs_condense_min_mapping_bytes 131072
zfs_dbgmsg_enable 1
zfs_dbgmsg_maxsize 4194304
zfs_dbuf_state_index 0
zfs_ddt_data_is_special 1
zfs_deadman_checktime_ms 60000
zfs_deadman_enabled 1
zfs_deadman_failmode wait
zfs_deadman_synctime_ms 600000
zfs_deadman_ziotime_ms 300000
zfs_dedup_prefetch 0
zfs_default_bs 9
zfs_default_ibs 17
zfs_delay_min_dirty_percent 60
zfs_delay_scale 500000
zfs_delete_blocks 20480
zfs_dirty_data_max 68719476736
zfs_dirty_data_max_max 180388626432
zfs_dirty_data_max_max_percent 25
zfs_dirty_data_max_percent 10
zfs_dirty_data_sync_percent 20
zfs_disable_ivset_guid_check 0
zfs_dmu_offset_next_sync 1
zfs_embedded_slog_min_ms 64
zfs_expire_snapshot 300
zfs_fallocate_reserve_percent 110
zfs_flags 0
zfs_free_bpobj_enabled 1
zfs_free_leak_on_eio 0
zfs_free_min_time_ms 1000
zfs_history_output_max 1048576
zfs_immediate_write_sz 32768
zfs_initialize_chunk_size 1048576
zfs_initialize_value 16045690984833335022
zfs_keep_log_spacemaps_at_export 0
zfs_key_max_salt_uses 400000000
zfs_livelist_condense_new_alloc 0
zfs_livelist_condense_sync_cancel 0
zfs_livelist_condense_sync_pause 0
zfs_livelist_condense_zthr_cancel 0
zfs_livelist_condense_zthr_pause 0
zfs_livelist_max_entries 500000
zfs_livelist_min_percent_shared 75
zfs_lua_max_instrlimit 100000000
zfs_lua_max_memlimit 104857600
zfs_max_async_dedup_frees 100000
zfs_max_log_walking 5
zfs_max_logsm_summary_length 10
zfs_max_missing_tvds 0
zfs_max_nvlist_src_size 0
zfs_max_recordsize 16777216
zfs_metaslab_find_max_tries 100
zfs_metaslab_fragmentation_threshold 70
zfs_metaslab_max_size_cache_sec 3600
zfs_metaslab_mem_limit 25
zfs_metaslab_segment_weight_enabled 1
zfs_metaslab_switch_threshold 2
zfs_metaslab_try_hard_before_gang 0
zfs_mg_fragmentation_threshold 95
zfs_mg_noalloc_threshold 0
zfs_min_metaslabs_to_flush 1
zfs_multihost_fail_intervals 0
zfs_multihost_history 1000
zfs_multihost_import_intervals 20
zfs_multihost_interval 1000
zfs_multilist_num_sublists 0
zfs_no_scrub_io 0
zfs_no_scrub_prefetch 0
zfs_nocacheflush 0
zfs_nopwrite_enabled 1
zfs_object_mutex_size 64
zfs_obsolete_min_time_ms 500
zfs_override_estimate_recordsize 0
zfs_pd_bytes_max 52428800
zfs_per_txg_dirty_frees_percent 30
zfs_prefetch_disable 0
zfs_read_history 0
zfs_read_history_hits 0
zfs_rebuild_max_segment 1048576
zfs_rebuild_scrub_enabled 1
zfs_rebuild_vdev_limit 67108864
zfs_reconstruct_indirect_combinations_max 4096
zfs_recover 0
zfs_recv_queue_ff 20
zfs_recv_queue_length 16777216
zfs_recv_write_batch_size 1048576
zfs_removal_ignore_errors 0
zfs_removal_suspend_progress 0
zfs_remove_max_segment 16777216
zfs_resilver_disable_defer 0
zfs_resilver_min_time_ms 3000
zfs_scan_blkstats 0
zfs_scan_checkpoint_intval 7200
zfs_scan_fill_weight 3
zfs_scan_ignore_errors 0
zfs_scan_issue_strategy 0
zfs_scan_legacy 0
zfs_scan_max_ext_gap 2097152
zfs_scan_mem_lim_fact 15
zfs_scan_mem_lim_soft_fact 20
zfs_scan_report_txgs 0
zfs_scan_strict_mem_lim 0
zfs_scan_suspend_progress 0
zfs_scan_vdev_limit 16777216
zfs_scrub_min_time_ms 1000
zfs_send_corrupt_data 0
zfs_send_no_prefetch_queue_ff 20
zfs_send_no_prefetch_queue_length 1048576
zfs_send_queue_ff 20
zfs_send_queue_length 16777216
zfs_send_unmodified_spill_blocks 1
zfs_slow_io_events_per_second 20
zfs_spa_discard_memory_limit 16777216
zfs_special_class_metadata_reserve_pct 25
zfs_sync_pass_deferred_free 2
zfs_sync_pass_dont_compress 8
zfs_sync_pass_rewrite 2
zfs_sync_taskq_batch_pct 75
zfs_traverse_indirect_prefetch_limit 32
zfs_trim_extent_bytes_max 134217728
zfs_trim_extent_bytes_min 32768
zfs_trim_metaslab_skip 0
zfs_trim_queue_limit 10
zfs_trim_txg_batch 32
zfs_txg_history 100
zfs_txg_timeout 5
zfs_unflushed_log_block_max 131072
zfs_unflushed_log_block_min 1000
zfs_unflushed_log_block_pct 400
zfs_unflushed_log_txg_max 1000
zfs_unflushed_max_mem_amt 1073741824
zfs_unflushed_max_mem_ppm 1000
zfs_unlink_suspend_progress 0
zfs_user_indirect_is_special 1
zfs_vdev_aggregate_trim 0
zfs_vdev_aggregation_limit 1048576
zfs_vdev_aggregation_limit_non_rotating 131072
zfs_vdev_async_read_max_active 3
zfs_vdev_async_read_min_active 1
zfs_vdev_async_write_active_max_dirty_percent 60
zfs_vdev_async_write_active_min_dirty_percent 30
zfs_vdev_async_write_max_active 10
zfs_vdev_async_write_min_active 2
zfs_vdev_cache_bshift 16
zfs_vdev_cache_max 16384
zfs_vdev_cache_size 0
zfs_vdev_default_ms_count 200
zfs_vdev_default_ms_shift 29
zfs_vdev_initializing_max_active 1
zfs_vdev_initializing_min_active 1
zfs_vdev_max_active 1000
zfs_vdev_max_auto_ashift 14
zfs_vdev_min_auto_ashift 9
zfs_vdev_min_ms_count 16
zfs_vdev_mirror_non_rotating_inc 0
zfs_vdev_mirror_non_rotating_seek_inc 1
zfs_vdev_mirror_rotating_inc 0
zfs_vdev_mirror_rotating_seek_inc 5
zfs_vdev_mirror_rotating_seek_offset 1048576
zfs_vdev_ms_count_limit 131072
zfs_vdev_nia_credit 5
zfs_vdev_nia_delay 5
zfs_vdev_open_timeout_ms 1000
zfs_vdev_queue_depth_pct 1000
zfs_vdev_raidz_impl cycle [fastest] original scalar sse2 ssse3 avx2 avx512f avx512bw
zfs_vdev_read_gap_limit 32768
zfs_vdev_rebuild_max_active 3
zfs_vdev_rebuild_min_active 1
zfs_vdev_removal_max_active 2
zfs_vdev_removal_min_active 1
zfs_vdev_scheduler unused
zfs_vdev_scrub_max_active 3
zfs_vdev_scrub_min_active 1
zfs_vdev_sync_read_max_active 10
zfs_vdev_sync_read_min_active 10
zfs_vdev_sync_write_max_active 10
zfs_vdev_sync_write_min_active 10
zfs_vdev_trim_max_active 2
zfs_vdev_trim_min_active 1
zfs_vdev_write_gap_limit 4096
zfs_vnops_read_chunk_size 1048576
zfs_wrlog_data_max 137438953472
zfs_zevent_len_max 512
zfs_zevent_retain_expire_secs 900
zfs_zevent_retain_max 2000
zfs_zil_clean_taskq_maxalloc 1048576
zfs_zil_clean_taskq_minalloc 1024
zfs_zil_clean_taskq_nthr_pct 100
zil_maxblocksize 131072
zil_min_commit_timeout 5000
zil_nocacheflush 0
zil_replay_disable 0
zil_slog_bulk 786432
zio_deadman_log_all 0
zio_dva_throttle_enabled 0
zio_requeue_io_start_cut_in_line 1
zio_slow_io_ms 30000
zio_taskq_batch_pct 80
zio_taskq_batch_tpq 0
zvol_inhibit_dev 0
zvol_major 230
zvol_max_discard_blocks 16384
zvol_prefetch_bytes 131072
zvol_request_sync 0
zvol_threads 32
zvol_volmode 1

VDEV cache disabled, skipping section

ZIL committed transactions: 0
Commit requests: 10
Flushes to stable storage: 10
Transactions to SLOG storage pool: 0 Bytes 0
Transactions to non-SLOG storage pool: 0 Bytes 0

We've done more testing and gathered more information for your review:

As a starter  Version of ZFS and Lustre required to reproduce the OOM:

ZFS version.   : 2.1.14_1llnl-1
Lustre Version: lustre-2.15.4_1.llnl
Ram total available: 187 Gib

FIRST RUN:

zfs_arc_max was set to default: 0

I also set the kernel to slab_nomerge to pinpoint the culprit slab if any.
There was no culprit slab found as nothing was jumping at the top of the in slabtop.
But we also check arcstat to see what was going on in the arc during testing.

command used arcstat 1:

 time  read  miss  miss%  dmis  dm%  pmis  pm%  mmis  mm%  size     c  avail
11:29:21   19K  6.8K     34   569    4  6.3K   99     0    0   93G   93G    25G
11:29:22   19K  6.9K     34   550    4  6.3K   99     0    0   93G   93G    25G
11:29:23   20K  7.0K     34   546    3  6.5K   99     0    0   93G   93G    25G
11:29:24   20K  7.2K     34   600    4  6.6K   99     0    0   93G   93G    25G
11:29:25   21K  7.5K     35   633    4  6.9K   99     0    0   94G   93G    24G
11:29:26   21K  7.6K     34   602    4  7.0K   99     0    0   93G   93G    25G
11:29:27   21K  7.8K     36   633    4  7.1K   99     0    0   94G   93G    25G
11:29:28   22K  8.0K     35   616    4  7.4K   99     0    0   94G   93G    24G
11:29:29   23K  9.0K     38  1.2K    7  7.8K   99   503    3   94G   93G    24G
11:29:30   22K   10K     45  2.7K   17  7.8K   99  2.0K   13   95G   93G    23G
11:29:31   23K   10K     46  2.7K   17  8.2K  100  2.0K   13   97G   93G    21G
11:29:32   24K   11K     46  2.8K   17  8.3K  100  2.1K   13   99G   93G    20G
11:29:33   24K   11K     46  2.8K   17  8.8K  100  2.1K   13  101G   93G    18G
11:29:34   24K   11K     46  2.7K   17  8.7K  100  2.0K   13  103G   93G    16G
11:29:35   26K   12K     47  2.9K   17  9.5K  100  2.1K   13  105G   93G    13G
11:29:36   26K   12K     47  2.8K   16  9.6K  100  2.0K   12  108G   93G    11G
11:29:37   26K   12K     47  2.8K   16  9.7K  100  2.0K   12  110G   93G   8.7G
11:29:38   27K   13K     47  2.8K   16   10K  100  2.0K   12  113G   93G   5.9G
11:29:39   27K   13K     47  2.9K   16   10K  100  2.0K   12  116G   93G   3.1G
11:29:40   27K   13K     48  2.7K   16   10K  100  1.9K   12  118G   93G    10M
11:29:41   41K   15K     37  5.7K   17  9.9K  100  4.9K   15  121G   93G  -3.6G

At that point we were OOMed.
--------------------------------------------------------------------------------------------------------------------------

SECOND RUN:

for the second run we set the zfs_arc_max to 47 Gib

keeping monitoring the arcstat we can see it going right through the limit set to 47Gib:

arcstat 1:

    time  read  miss  miss%  dmis  dm%  pmis  pm%  mmis  mm%  size     c  avail
11:57:19   19K  9.0K     47  3.1K   23  5.9K  100  2.5K   20   76G   47G    41G
11:57:20   19K  9.1K     47  3.1K   23  6.1K  100  2.5K   19   77G   47G    40G
11:57:21   20K  9.4K     47  3.0K   22  6.4K  100  2.4K   18   78G   47G    39G
11:57:22   20K  9.6K     47  3.1K   22  6.5K  100  2.5K   18   79G   47G    38G
11:57:23   21K  10.0K     47  3.2K   22  6.8K  100  2.5K   18   80G   47G    37G
11:57:24   21K  10.0K     47  3.1K   21  6.9K  100  2.4K   17   81G   47G    36G
11:57:25   21K   10K     47  3.2K   21  7.2K  100  2.5K   18   82G   47G    35G
11:57:26   22K   10K     47  3.2K   21  7.4K  100  2.5K   17   84G   47G    34G
11:57:27   22K   10K     47  3.0K   20  7.8K  100  2.4K   16   85G   47G    32G
11:57:28   23K   11K     47  3.2K   20  8.0K  100  2.5K   16   87G   47G    30G
11:57:29   23K   11K     47  3.1K   19  8.3K  100  2.4K   16   89G   47G    29G
11:57:30   24K   11K     48  3.2K   20  8.7K  100  2.5K   16   91G   47G    27G
11:57:31   24K   12K     48  3.2K   19  8.9K  100  2.5K   16   93G   47G    24G
11:57:32   25K   12K     48  3.1K   18  9.4K  100  2.4K   15   95G   47G    22G
11:57:33   26K   12K     48  3.2K   19  9.6K  100  2.4K   15   98G   47G    20G
11:57:34   26K   13K     49  3.2K   19  9.9K  100  2.5K   15  100G   47G    17G
11:57:35   27K   13K     49  3.3K   18   10K  100  2.5K   15  103G   47G    14G
11:57:36   28K   14K     49  3.4K   19   10K  100  2.6K   15  106G   47G    11G
11:57:37   28K   14K     49  3.3K   18   11K  100  2.5K   14  109G   47G   8.4G
11:57:38   29K   14K     50  3.3K   18   11K  100  2.6K   14  113G   47G   5.0G
11:57:39   30K   15K     50  3.3K   17   11K  100  2.5K   14  116G   47G   669M
11:57:40   44K   17K     39  6.1K   18   11K  100  5.3K   16  120G   47G  -3.9G

------------------------------------------------------------------------------------------------------------

I will look into zfs with our zfs developers and update ticket.

Originally I thought this was related to cgroups, which is a client side issue, but I didn't notice the "OSS" in the summary.

The majority of memory usage looks to be in "slab_reclaimable:185039 slab_unreclaimable:2082954" or at least I can't see anything else reported in the meminfo dump.  Are you able to capture /proc/slabinfo or slabtop from the OSS while the IOR is running, and see what is using the majority of memory?

This might relate to the use of deferred fput on the server, which can accumulate over time if the server was running a long time? There were two recent patches related to this that landed on master, but these may only be relevant for osd-ldiskfs and not osd-zfs (which I assume is the case here):

• https://review.whamcloud.com/51731 "LU-16973 osd: adds SB_KERNMOUNT flag"
• https://review.whamcloud.com/51805 "LU-16973 ptlrpc: flush delayed file desc if idle"

carbonneau, ofaaland,

it would be useful to include the actual stack traces from the OSS when the OOM is hit, not just the meminfo.  Otherwise it is difficult to know what is actually allocating the memory.  Sometimes it is just an innocent bystander process, but in many cases the actual offender is caught because it is the one allocating memory the most frequently...

I forgot to mention the issue occurs during read operations on the OSS. During the write operations the OSS memory was consnt.

Thanks, Peter. In our case, when we reproduced this, the client was 2.12 and the server was 2.14 or 2.15.

Are you saying client patches might fix this? We're happy to test master clients, but I would think the server should be managing its memory usage without depending on the client to behave in a certain way.

Eric

There have been some recent changes merged to master for the upcoming 2.16 release that we think could well help address this problem. Could you please retry your reproducer against a master client. If that does indeed resolve the issue then we can look to what would need to be back ported to b2_15 in order to get the same benefit there

Regards

Peter