Co-routine stack size check is failure prone

[ericonr]

Bug Report

Describe the bug

As seen in #2464, launching fluent-bit makes it spit out this error message:

Error: Invalid coroutine stack size. Aborting

This happens due to

fluent-bit/src/fluent-bit.c

Line 1111 in 5ffc991

if (config->coro_stack_size < getpagesize()) {

and

fluent-bit/include/fluent-bit/flb_coro.h

Line 68 in 5ffc991

#define FLB_CORO_STACK_SIZE ((3 * PTHREAD_STACK_MIN) / 2)

musl defines PTHREAD_STACK_MIN to 2048, which means that the result of PTHREAD_STACK_MIN * 3 / 2 is 3072, which is definitely smaller than what getpagesize() returns (4096 on most platforms). I'm not sure what the best solution here is, either don't use getpagesize as the basis for the check, or use the maximum between getpagesize and PTHREAD_STACK_MIN * 3 / 2.

This also fails on platforms that support 64k pages (aarch64 and ppc64), even when using glibc. Most aarch64 machines won't be configured that way, but to my knowledge all ppc64 distros besides Void Linux use 64k pages.

To Reproduce

Simply try to launch fluent-bit when built against musl libc.

Expected behavior

It will at least launch.

Your Environment

• Version used: built from master (305590f)
• Operating System and version: any musl based distro

[bgdnlp]

The same thing happens on FreeBSD. Running just fluent-bit immediately complains that

Fluent Bit v1.7.9
* Copyright (C) 2019-2021 The Fluent Bit Authors
* Copyright (C) 2015-2018 Treasure Data
* Fluent Bit is a CNCF sub-project under the umbrella of Fluentd
* https://fluentbit.io

Error: Invalid coroutine stack size. Aborting

fluent-bit --help says:

  -s, --coro_stack_size   set coroutines stack size in bytes (default: 3072)

Would it be acceptable to simply set coro_stack_size to getpagesize() if the default is smaller and it isn't otherwise configured?

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[ericonr]

I would ask of the repo maintainers to remove the auto stale bot. I have nothing to do with upstream not answering this issue yet. If the bot pings the issue again I will just close it and give up on the bug report.

[girgen]

It seems stupid that the binary just fails to run unless you set the coro_stack_size?

I maintain the FreeBSD port, and I'm patching the source to double the stack size, in wait for a better fix.

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[bgdnlp]

Could someone please have a look at this issue, maybe it's not controversial and can be fixed easily?

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[bgdnlp]

Aaaand, bump

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[krystiannowak]

another bump

[edsiper]

note: we are adjusting the auto-stale bot, quite noisy.

assigning @nokute78 to review this.

[nokute78]

Would it be acceptable to simply set coro_stack_size to getpagesize() if the default is smaller and it isn't otherwise configured?

I agree.

I sent a patch #4434 to change the default size if it is smaller than page size.
User can overwrite via config/option.