This message was deleted.
# generals
Slackbot
10/25/2023, 9:21 AMThis message was deleted.
y
Yu Wei Sung
10/25/2023, 12:33 PMDo you have metrics setup on bookies? My first guess is on bookie side. In the beginning, it is fast because the messages are still in cache.
👍 1
m
miton18
10/25/2023, 12:35 PMYes, we have metrics on bookies, have you one in mind ?
a
Asaf Mesika
10/25/2023, 1:18 PM@Hang Chen You have a talk on this no?
l
Lari Hotari
10/25/2023, 2:27 PMI gave a talk last year that gives some advice.
Slides:
https://www.apachecon.com/acna2022/slides/03_Hotari_Lari_Performance_tuning_Pulsar.pdf
Recording:
m
miton18
10/25/2023, 3:02 PMThanks, I will take time to watch it
h
Hang Chen
10/26/2023, 9:30 AMFrom the logs, the p99 read latency is two high, and you’d better check the bookie dashboard
k
kannarfr
10/29/2023, 3:28 PMWe can see heavy latency on reads. We are running on 8 threads VM, with default BK configuration so maybe we are configuring too much threads for the VM resources but the CPU load/RAM usages are not heavy.
kannarfr
10/29/2023, 3:36 PM Copy code
# TYPE jvm_threads_current gauge
jvm_threads_current{} 106.0
# TYPE jvm_threads_daemon gauge
jvm_threads_daemon{} 12.0
# TYPE jvm_threads_peak gauge
jvm_threads_peak{} 108.0
# TYPE jvm_threads_started counter
jvm_threads_started_total{} 183.0
# TYPE jvm_threads_deadlocked gauge
jvm_threads_deadlocked{} 0.0
# TYPE jvm_threads_deadlocked_monitor gauge
jvm_threads_deadlocked_monitor{} 0.0
# TYPE jvm_threads_state gauge
jvm_threads_state{state="NEW"} 0.0
jvm_threads_state{state="TERMINATED"} 0.0
jvm_threads_state{state="RUNNABLE"} 46.0
jvm_threads_state{state="BLOCKED"} 0.0
jvm_threads_state{state="WAITING"} 47.0
jvm_threads_state{state="TIMED_WAITING"} 13.0
jvm_threads_state{state="UNKNOWN"} 0.0
# TYPE replication_bookkeeper_client_BookKeeperClientWorker_threads gauge
replication_bookkeeper_client_BookKeeperClientWorker_threads 8
# TYPE bookkeeper_server_BookieHighPriorityThread_threads gauge
bookkeeper_server_BookieHighPriorityThread_threads 8
# TYPE bookkeeper_server_BookieReadThreadPool_threads gauge
bookkeeper_server_BookieReadThreadPool_threads 16kannarfr
10/29/2023, 4:11 PM45K context switch per second is legit?
y
Yu Wei Sung
10/30/2023, 12:49 PMhigh cs is normal as long as sys is not over single digit.
k
kannarfr
11/03/2023, 12:06 AMI have a lot of auth/disconnections from brokers such as
Copy code
Nov 03 00:06:19 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:19,176+0000 [bookie-io-8-61] INFO org.apache.bookkeeper.proto.AuthHandler - Authentication success on server side
Nov 03 00:06:19 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:19,176+0000 [bookie-io-8-61] INFO org.apache.bookkeeper.proto.BookieRequestHandler - Channel connected [id: 0x72e0b545, L:/192.168.4.1:3181 - R:/192.168.4.14:53946]
Nov 03 00:06:19 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:19,251+0000 [bookie-io-8-57] INFO org.apache.bookkeeper.proto.BookieRequestHandler - Channels disconnected: [id: 0xd092da68, L:/192.168.4.1:3181 ! R:/192.168.4.2:34750]
Nov 03 00:06:19 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:19,722+0000 [bookie-io-8-62] INFO org.apache.bookkeeper.proto.AuthHandler - Authentication success on server side
Nov 03 00:06:19 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:19,722+0000 [bookie-io-8-62] INFO org.apache.bookkeeper.proto.BookieRequestHandler - Channel connected [id: 0x2228a62c, L:/192.168.4.1:3181 - R:/192.168.4.25:42920]
Nov 03 00:06:20 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:20,102+0000 [bookie-io-8-63] INFO org.apache.bookkeeper.proto.AuthHandler - Authentication success on server side
Nov 03 00:06:20 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:20,103+0000 [bookie-io-8-63] INFO org.apache.bookkeeper.proto.BookieRequestHandler - Channel connected [id: 0x55aae591, L:/192.168.4.1:3181 - R:/192.168.4.3:39146]
Nov 03 00:06:20 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:20,566+0000 [bookie-io-8-64] INFO org.apache.bookkeeper.proto.AuthHandler - Authentication success on server side
Nov 03 00:06:20 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:20,566+0000 [bookie-io-8-64] INFO org.apache.bookkeeper.proto.BookieRequestHandler - Channel connected [id: 0x9602b7e2, L:/192.168.4.1:3181 - R:/192.168.4.22:34308]
Nov 03 00:06:20 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:20,864+0000 [bookie-io-8-65] INFO org.apache.bookkeeper.proto.AuthHandler - Authentication success on server side
Nov 03 00:06:20 clevercloud-bookkeeper-c3-n1 pulsar[713846]: 2023-11-03T00:06:20,864+0000 [bookie-io-8-65] INFO org.apache.bookkeeper.proto.BookieRequestHandler - Channel connected [id: 0x49c37bce, L:/192.168.4.1:3181 - R:/192.168.4.20:44160]
^Ckannarfr
11/03/2023, 12:07 AMis that normal behavior ?
kannarfr
11/03/2023, 12:07 AMalso, it looks BK doesn't check for epoll availability
kannarfr
11/03/2023, 12:09 AMI think I have a lot of epoll.wait ongoing concurrently
kannarfr
11/03/2023, 2:50 PMWe have good writes rates, only reads are very low.
kannarfr
11/03/2023, 3:44 PM Copy code
yo-bookkeeper-c3-n31 ~ # curl <http://localhost:9102/metrics> -s | grep SCHEDULING
# TYPE bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY summary
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="false",quantile="0.5"} NaN
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="false",quantile="0.75"} NaN
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="false",quantile="0.95"} NaN
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="false",quantile="0.99"} NaN
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="false",quantile="0.999"} NaN
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="false",quantile="0.9999"} NaN
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="false",quantile="1.0"} -Infinity
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY_count{success="false"} 0
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY_sum{success="false"} 0.0
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="true",quantile="0.5"} 224.811
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="true",quantile="0.75"} 590.511
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="true",quantile="0.95"} 9647.003
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="true",quantile="0.99"} 14450.689
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="true",quantile="0.999"} 15079.686
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="true",quantile="0.9999"} 15205.121
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY{success="true",quantile="1.0"} 15322.502
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY_count{success="true"} 39986424
bookkeeper_server_READ_ENTRY_SCHEDULING_DELAY_sum{success="true"} 4.1107050174E10l
Lari Hotari
11/03/2023, 3:51 PMWe have good writes rates, only reads are very low.
Interesting. I guess you haven't found the bottleneck yet? Could you check the RocksDB entry location index lookups metrics? They were added in this PR: https://github.com/apache/bookkeeper/pull/3444 .
k
kannarfr
11/03/2023, 3:52 PMIndeed. Ok thanks, checking that.
kannarfr
11/03/2023, 3:55 PMI'll bump BK from 4.16.3 to 4.16.6.
kannarfr
11/03/2023, 3:59 PM@Lari Hotari our use cases is ~15000topics having ~1000msg/s, on we do reads on these from custom messageId. Any RocksDB tuning advices?
l
Lari Hotari
11/03/2023, 4:06 PMWhat I would do is first trying to understand where the bottleneck is then experiment with tuning that the bottleneck can be eliminated. For RocksDB, you need to give enough RAM for it so that lookups happen in memory and don't cause disk IO. What's your current bookkeeper config?
Lari Hotari
11/03/2023, 4:10 PMdbStorage_rocksDB_blockCacheSizek
kannarfr
11/03/2023, 4:10 PMWe use the default one.
kannarfr
11/03/2023, 4:11 PMWe tried using BK on
-Xms8192m -Xmx8192m -XX:MaxDirectMemorySize=8076mkannarfr
11/03/2023, 4:11 PMAnd
-Xms24g -Xmx24g -XX:MaxDirectMemorySize=24gkannarfr
11/03/2023, 4:13 PM(With a 16GB RAM, and a 64GB RAM)
kannarfr
11/03/2023, 4:14 PMBTW, we put too much Xmx, MaxdirectMemorySize for the 16GB 😮. Maybe it's the root cause. Fixing..
l
Lari Hotari
11/03/2023, 4:17 PMbtw.
dbStorage_rocksDB_blockCacheSizek
kannarfr
11/03/2023, 4:17 PMAck
l
Lari Hotari
11/03/2023, 4:18 PMit could be less than 10% when you have multiple ledger directories: https://github.com/apache/bookkeeper/blob/ad0ed213c0d3abf971441c7160334af99d94159c[…]pache/bookkeeper/bookie/storage/ldb/KeyValueStorageRocksDB.java
y
Yu Wei Sung
11/03/2023, 4:30 PMdo you have worker node iostat?
l
Lari Hotari
11/03/2023, 4:39 PMour use cases is ~15000topics having ~1000msg/s, on we do reads on these from custom messageId. Any RocksDB tuning advices?@kannarfr when you say "we do reads on these from custom messageId", do you mean that you start readers from a specific messageId as the earliest message? what is the approximate rate of this type of workload? how many messages are typically read using the reader before it is closed? Just trying to understand the usecase and the access pattern that it has.
k
kannarfr
11/03/2023, 4:52 PMvdb is used as storage for BK:
Copy code
yo-bookkeeper-c3-n1 ~ # iostat
Linux 6.1.19+ (yo-bookkeeper-c3-n1) 03/11/23 _x86_64_ (50 CPU)
avg-cpu: %user %nice %system %iowait %steal %idle
3.46 0.00 1.95 0.47 0.02 94.10
Device tps kB_read/s kB_wrtn/s kB_dscd/s kB_read kB_wrtn kB_dscd
vda 34.13 45.85 181.00 0.00 4345754 17154144 0
vdb 1760.44 1178.39 19142.06 0.00 111684161 1814225152 0kannarfr
11/03/2023, 4:52 PMIn fact isn't really 1000msg/s. We have thousands of applications that push their logs (one topic per app).
kannarfr
11/03/2023, 4:52 PMThen depending on requests we need to provide logs for an application.
kannarfr
11/03/2023, 4:53 PMWe use the admin api to get the messageid for a given timestamp and we start a reader for it.
l
Lari Hotari
11/03/2023, 4:53 PMwhat type of disks do you have? local or over network? SSD or HDD?
k
kannarfr
11/03/2023, 4:54 PMEach BK has its own SSD directly assign through qemu.
kannarfr
11/03/2023, 4:54 PMEach BK has a 4TB SSD
l
Lari Hotari
11/03/2023, 4:55 PMlocal SSD, so no network storage system (SAN)?
k
kannarfr
11/03/2023, 4:55 PM Copy code
yo-bookkeeper-c3-n1 /data # dd if=/dev/zero of=test bs=20G count=1 oflag=dsync
0+1 records in
0+1 records out
2147479552 bytes (2.1 GB, 2.0 GiB) copied, 2.78952 s, 770 MB/skannarfr
11/03/2023, 4:55 PMYes local SSD.
👍 1
kannarfr
11/03/2023, 4:58 PMWe have 5 proxies/ 5brokers, and 34 BK nodes as mentioned.
kannarfr
11/03/2023, 4:58 PMShared in 3 DC with regionaware
l
Lari Hotari
11/03/2023, 5:04 PMit feels that there's sufficient amount of HW resources for the use case. 🙂 it will definitely help to find the metrics which pinpoint the problem. My current assumption is that the Rocksdb entry location lookups could be a bottleneck. The metrics could reveal that. https://github.com/apache/bookkeeper/commit/19fd8f74 is already in 4.16.0 so there should be a need to bump BK version to get the metrics if you are already on 4.16.x.
One way to experiment with dbStorage_rocksDB_blockCacheSize tuning would be to set it to the 2GB in bytes which is
2147483648k
kannarfr
11/03/2023, 5:24 PMOk thanks, I'll keep you posted!
kannarfr
11/03/2023, 8:30 PMWe applied the 2GB to blockCacheSize, but maybe we also need to add more RAM/cpu (on vm, so jvm) and threads (read, write, journal etc. in bk conf).
And here is a representative metrics of lookup_entry_location for one BK node:
Copy code
bookie_lookup_entry_location{success="false",quantile="0.5", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.535
bookie_lookup_entry_location{success="false",quantile="0.75", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.535
bookie_lookup_entry_location{success="false",quantile="0.95", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.535
bookie_lookup_entry_location{success="false",quantile="0.99", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.535
bookie_lookup_entry_location{success="false",quantile="0.999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.535
bookie_lookup_entry_location{success="false",quantile="0.9999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.535
bookie_lookup_entry_location{success="false",quantile="1.0", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.535
bookie_lookup_entry_location_count{success="false", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 202
bookie_lookup_entry_location_sum{success="false", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 2842.0
bookie_lookup_entry_location{success="true",quantile="0.5", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 10.578
bookie_lookup_entry_location{success="true",quantile="0.75", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 12.398
bookie_lookup_entry_location{success="true",quantile="0.95", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 14.001
bookie_lookup_entry_location{success="true",quantile="0.99", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 15.024
bookie_lookup_entry_location{success="true",quantile="0.999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 17.495
bookie_lookup_entry_location{success="true",quantile="0.9999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 22.982
bookie_lookup_entry_location{success="true",quantile="1.0", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 25.474
bookie_lookup_entry_location_count{success="true", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 991563
bookie_lookup_entry_location_sum{success="true", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 1.0882591E7kannarfr
11/03/2023, 8:32 PMMachines with 8threads (vCPU), AMD EPYC 7713P, 15GB RAMavailable : 4GB xmx/xms, 4GB maxdirectmemorysize and 2GB blockcachesize
kannarfr
11/03/2023, 8:33 PM Copy code
bookie_lookup_entry_location{success="false",quantile="0.5", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} NaN
bookie_lookup_entry_location{success="false",quantile="0.75", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} NaN
bookie_lookup_entry_location{success="false",quantile="0.95", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} NaN
bookie_lookup_entry_location{success="false",quantile="0.99", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} NaN
bookie_lookup_entry_location{success="false",quantile="0.999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} NaN
bookie_lookup_entry_location{success="false",quantile="0.9999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} NaN
bookie_lookup_entry_location{success="false",quantile="1.0", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} -Infinity
bookie_lookup_entry_location_count{success="false", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 45
bookie_lookup_entry_location_sum{success="false", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 626.0
bookie_lookup_entry_location{success="true",quantile="0.5", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 6.934
bookie_lookup_entry_location{success="true",quantile="0.75", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 9.568
bookie_lookup_entry_location{success="true",quantile="0.95", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 10.465
bookie_lookup_entry_location{success="true",quantile="0.99", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 11.192
bookie_lookup_entry_location{success="true",quantile="0.999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 12.949
bookie_lookup_entry_location{success="true",quantile="0.9999", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 12.949
bookie_lookup_entry_location{success="true",quantile="1.0", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 15.896
bookie_lookup_entry_location_count{success="true", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 634044
bookie_lookup_entry_location_sum{success="true", indexDir="/data/bookkeeper/ledgers/current",ledgerDir="/data/bookkeeper/ledgers/current"} 6079337.0l
Lari Hotari
11/04/2023, 11:58 AM@kannarfr what are E, Qw and Qa values? The reason I'm asking is that sticky reads aren't enabled in the bookkeeper client unless ensemble size equals write quorum size. However, you should also be able to see that from metrics when cache misses are high in bookkeeper. without sticky reads, the read ahead cache won't be efficient in bk.
k
kannarfr
11/04/2023, 11:59 AME=5, Qw3, Qa=2
kannarfr
11/04/2023, 11:59 AMOk, bumping conf.. 🙂
l
Lari Hotari
11/04/2023, 12:00 PMLari Hotari
11/04/2023, 12:00 PMgot burned by that before... 🙂 that's why I made https://github.com/apache/pulsar/pull/18003
k
kannarfr
11/04/2023, 12:01 PMClear, thanks a lot!
👍 1
l
Lari Hotari
11/04/2023, 12:05 PMWhen you have a lot of consumers, there are some gotchas in tuning the broker entry cache which makes a big difference. In your use case, I guess you don't have a lot of consumers that could benefit from a well tuned broker entry cache?
k
kannarfr
11/04/2023, 12:06 PMTalking about
dispatcherMaxReadBatchSizel
Lari Hotari
11/04/2023, 12:08 PMno,
managedLedgerCache*managedLedger*ForCachingk
kannarfr
11/04/2023, 12:09 PMI think it can be worth it to increase them. We can have several consumers created on the same topic when we looking for errors in logs, like read part of topic, search reread with filters etc.
kannarfr
11/04/2023, 12:09 PMNice tip 🙂.
l
Lari Hotari
11/04/2023, 12:10 PMtuning those help tailing read and catchup read scenarios with a large amount of consumers (100s / 1000s)
k
kannarfr
11/04/2023, 12:10 PMClear, ty
l
Lari Hotari
11/04/2023, 12:14 PMWhich Pulsar version are you using?
k
kannarfr
11/04/2023, 12:15 PM3.1.1
👍 1
kannarfr
11/06/2023, 3:07 PMIs it safe when cluster is already running to restart bk with the
entryLogPerLedgerEnabled=truekannarfr
11/06/2023, 3:09 PM(how many open files can we tolerate for it?)
l
Lari Hotari
11/07/2023, 8:11 AMI don't remember touching that config. A search in the source code reveals that there's a need to set
useTransactionalCompaction=trueentryLogPerLedgerEnabled=truek
kannarfr
11/08/2023, 11:45 PMFYI, since ensemble=Qw, reads rates have increased x30. We will soon increase broker readbatchsize (on bk reads) from 100 to 500 and see if it increases the rates too (it won't be that big but let's see).
kannarfr
11/08/2023, 11:45 PMThanks a lot for your diag & advices Lari!
thankyou 1
a
Asaf Mesika
11/09/2023, 9:28 AM@kannarfr Help you fellow users and include this tip in the docs . Very easy to do PR for pulsar-site repo
l
Lari Hotari
11/09/2023, 9:49 AM@Asaf Mesika that was my intention with the PR referenced in https://apache-pulsar.slack.com/archives/C5Z4T36F7/p1699099254267459?thread_ts=1698225686.705339&cid=C5Z4T36F7
a
Asaf Mesika
11/09/2023, 11:21 AMHmm @kannarfr where would you put if, for your last-week self for you to notice it?
k
kannarfr
11/09/2023, 3:04 PMMaybe a bullet point about stickyreads here https://bookkeeper.apache.org/docs/admin/bookies#performance?
💯 1
a
Asaf Mesika
11/13/2023, 10:56 AM@kannarfr Would you like to contribute the fix to this section?
k
kannarfr
11/15/2023, 2:31 PMhttps://github.com/apache/bookkeeper/pull/4131, as it's public doc, maybe you have some hints about phrasing. Do not hesitate to tell me.
3 Views