Parser Filter on Docker reduces Fluent Bit benchmark performance from 100MB/s to 15MB/s

[mselinger75]

Bug Report

Describe the bug
I have a log generator which produces 30mb/sec logs to the file. Then I have FluentBit setup to tail this file, and send to NULL output.
When I look at FluentBit metrics I see that number of logs processed by tail input is way less than number of logs generated.
I've tried with different log volumes:
100mb/sec, 50mb/sec, 30mb/sec, 15mb/sec. The only rate I see FluentBit is able to keep up with is 15mb/sec.

I tested on versions 1.9.8 and 2.1.6, for 2.1.6 I've tried threaded on - didn't help.
For 2.1.6 when I set threaded on - I see a lot of errors in FluentBit log:

[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0

To Reproduce
My config:

[SERVICE]
    Flush         1
    Log_Level     info
    Daemon        off
    Parsers_File  parsers.conf
    HTTP_Server   On
    HTTP_Listen   0.0.0.0
    HTTP_Port     2020
    storage.metrics            on
    storage.sync               normal
    storage.checksum           off
    storage.max_chunks_up      128
    storage.backlog.mem_limit  100M
    storage.path               /var/log/flb/storage/

[FILTER]
    Name parser
    Match k8s_audit
    Key_Name log
    Parser docker

[INPUT]
    Name               tail
    Tag                k8s_audit
    Path               /usr/local/home/test/fluentbit/output.txt
    DB                 /var/log/flb/fluent-bit.db
    DB.locking         true
    Read_from_Head     false
    Buffer_Max_Size    20M
    Buffer_Chunk_Size  2MB
    Mem_Buf_Limit      100MB
    Refresh_Interval   5
    #Rotate_Wait 15
    storage.type       filesystem
    threaded on
    Skip_Long_Lines    On

[Output]
    Name null
    Match k8s_audit
    Retry_Limit               no_limits
    workers 8

• Steps to reproduce the problem:
  Run FluentBit with config above:
  bin/fluent-bit -c /usr/local/home/test/fluentbit/fluent-bit.conf

Generate logs to /usr/local/home/test/fluentbit/output.txt
Check that FluentBit metrics are behind the logs volume by looking at http://localhost:2020/api/v1/metrics

Expected behavior
FluentBit should be able to keep up with logs volume

Your Environment
Debian 6.1.25-1rodete1 (2023-05-11) x86_64 GNU/Linux

• Version used:
  Both 1.9.8 and 2.1.6

[agup006]

What application is writing to the filesystem? Also is it running on the same Operating System as where Fluent Bit is running? Lastly, what are the specifications of your server?

[mselinger75]

What application is writing to the filesystem?

https://github.com/GoogleCloudPlatform/ops-agent/blob/master/integration_test/soak_test/cmd/launcher/log_generator.py

is it running on the same Operating System as where Fluent Bit is running?

yes

what are the specifications of your server?

Architecture:            x86_64
  CPU op-mode(s):        32-bit, 64-bit
  Address sizes:         48 bits physical, 48 bits virtual
  Byte Order:            Little Endian
CPU(s):                  128
  On-line CPU(s) list:   0-127
Vendor ID:               AuthenticAMD
  Model name:            AMD EPYC 7B12


MemTotal:       528330084 kB
MemFree:        328793528 kB
MemAvailable:   494561172 kB

[agup006]

What disk is it using?

[mselinger75]

What disk is it using?

which command do you want me run for that?

Also, I forgot to mention: once I remove docker FILTER plugin from the FluentBit config - I can process 100mb/sec no problem

[agup006]

What disk is it using?

which command do you want me run for that?

No specific command, more if it is using SSD / HDD and if we are facing limitations of I/O. Based on your next comment though probably not relevant

Also, I forgot to mention: once I remove docker FILTER plugin from the FluentBit config - I can process 100mb/sec no problem

Yes very helpful :), what happens if you add the Docker Parser in the Input tail?

[mselinger75]

Yes very helpful :), what happens if you add the Docker Parser in the Input tail?

same issue

[mselinger75]

One more note:
For 2.1.6 when I set threaded on - metrics seems to match log generator numbers, but I see a lot of errors in FluentBit log:

[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0

not sure what does it mean

[agup006]

One more note: For 2.1.6 when I set threaded on - metrics seems to match log generator numbers, but I see a lot of errors in FluentBit log:

Could you elaborate on what you mean by metrics seem to match log generator numbers? Is this saying you are not seeing a performance bottleneck?

[tail.0] failed buffer write, retries=0 [tail.0] failed buffer write, retries=0 [tail.0] failed buffer write, retries=0 [tail.0] failed buffer write, retries=0 [tail.0] failed buffer write, retries=0 [tail.0] failed buffer write, retries=0 [tail.0] failed buffer write, retries=0

not sure what does it mean

[mselinger75]

For 2.1.6 when I set threaded on for 100mb/sec and run for few minutes I see:

1. Backpressure on file system.
2. In FluentBit logs I see

[2023/07/05 14:22:12] [ info] [input:tail:tail.0] inode=15370259 handle rotation(): /usr/local/home/fbit/output.txt => /usr/local/home/fbit/output.txt.old
[2023/07/05 14:22:12] [ info] [input:tail:tail.0] inotify_fs_remove(): inode=15370259 watch_fd=15
[2023/07/05 14:22:12] [ info] [input:tail:tail.0] inotify_fs_add(): inode=15370259 watch_fd=16 name=/usr/local/home/fbit/output.txt.old
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=1
[tail.0] failed buffer write, retries=2
[tail.0] failed buffer write, retries=3
[tail.0] failed buffer write, retries=4
[tail.0] failed buffer write, retries=5
[tail.0] failed buffer write, retries=6
[tail.0] failed buffer write, retries=7
[tail.0] failed buffer write, retries=8
[tail.0] failed buffer write, retries=9
[2023/07/05 14:22:16] [error] [input:tail:tail.0] could not enqueue records into the ring buffer
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=1
[tail.0] failed buffer write, retries=2
[tail.0] failed buffer write, retries=3
[tail.0] failed buffer write, retries=4
[tail.0] failed buffer write, retries=5
[tail.0] failed buffer write, retries=6
[tail.0] failed buffer write, retries=7
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[2023/07/05 14:22:19] [ info] [input:tail:tail.0] inotify_fs_add(): inode=15370063 watch_fd=17 name=/usr/local/home/fbit/output.txt
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[2023/07/05 14:22:20] [ info] [input:tail:tail.0] inotify_fs_remove(): inode=15370259 watch_fd=16
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0
[tail.0] failed buffer write, retries=0

3. When I stop log generator, I see backpressure resolved within ~1 min. When I check http://localhost:2020/api/v1/metrics I see number of records matches number of logs generated.

I also see fbit-pipeline thread consumes 100% of CPU:

On a side note: I noticed multiple time a problem when if I stop log generator, FluentBit also stops processing tail input, even though I see it didn't process all the records according to metrics stats. If I restart FluentBit I see it immediately processes the rest of the file with logs.

[lecaros]

Hi @mselinger75
could you please try with a build from this PR?
#7815

[github-actions]

This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 5 days. Maintainers can add the exempt-stale label.

[github-actions]

This issue was closed because it has been stalled for 5 days with no activity.