Add Fluent::Plugin::Base#after_start to detect end of #start #1190
Conversation
This is to make a detection that #start method completely finished. Without this method, flush/enqueue threads will run without consideration even if Output#start takes long time to finish. Output#start may need long time to finish if it tries network connection or service detection over internet.
|
@repeatedly Please check this change. This should be included in next release. |
| @@ -922,6 +927,11 @@ def enqueue_thread_run | |||
| interval = @buffer_config.flush_thread_interval | |||
| end | |||
|
|
|||
| while !self.after_started? && !self.stopped? | |||
| sleep interval | |||
There was a problem hiding this comment.
Is actual interval value needed for this check?
sleep 1 is not good?
There was a problem hiding this comment.
sleep 1 works well... right, longer interval is not good. I'll update code.
|
It's green on CI. I'll merge this later. |
|
Does this addition require plugins to call I was testing things using Fluentd v0.14.12 and the Elasticsearch output plugin v1.9.2 (which extends from As a result, this would eventually lead to a def start
super
self.after_start
end |
|
No. elasitcsearch plugin v1.9.2 uses v0.12 API and v0.14 uses compat layer for it. |
This method (and flag which shows this method is called properly) helps threads to start its actual process when all preparations are done.
This fixes #1187.