fix(pdk/vault): do not call sema:wait in init phase

[chronolaw]

Summary

Fix issue #9839

Checklist

• The Pull Request has tests
• There's an entry in the CHANGELOG

Full changelog

• check nginx phase
• if it is init or init_worker, set wait_ok to false

Issue reference

Fix #[9839]

[chronolaw]

I think it should have a change log entry, but I don't know how to describe it.
Could someone help me for it?

[mong0520]

hi @chronolaw ,

I'm the issue submitter of #9839.
First thank you for the fixing, and I have few questions about this PR.

1. Since sema:wait only works in some specific phases, how about set wait_ok to true if those phases are detected instead of set it to false in init false only?

# https://github.com/openresty/lua-resty-core/blob/master/lib/ngx/semaphore.md#wait
syntax: ok, err = sema:wait(timeout)
context: rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.*

2. Won't it be any problem if not calling to RETRY_SEMAPHORE:wait(RETRY_WAIT) but calls to RETRY_SEMAPHORE:post() in subsequence codes? I don't know if they should be paired.

[chronolaw]

hi @chronolaw ,

I'm the issue submitter of #9839. First thank you for the fixing, and I have few questions about this PR.

1. Since sema:wait only works in some specific phases, how about set wait_ok to true if those phases are detected instead of set it to false in init false only?

# https://github.com/openresty/lua-resty-core/blob/master/lib/ngx/semaphore.md#wait
syntax: ok, err = sema:wait(timeout)
context: rewrite_by_lua*, access_by_lua*, content_by_lua*, ngx.timer.*

2. Won't it be any problem if not calling to RETRY_SEMAPHORE:wait(RETRY_WAIT) but calls to RETRY_SEMAPHORE:post() in subsequence codes? I don't know if they should be paired.

1. In Kong, pdk code will run in init init_worker access rewrite phases, I think it is OK.
2. We set wait_ok = false, so subsequence code will not call RETRY_SEMAPHORE:post()

[bungle]

@chronolaw thank you!

[mong0520]

@chronolaw
One more question.
It looks like an always crash issue (calls to sema:wait() in init phase), but why there is no one reports it since 3.0.1 was released for a month?