feat: add fluent bit config-reload via HTTP #1286
Conversation
Signed-off-by: jiuxia211 <[email protected]>
| @@ -118,6 +118,8 @@ type Service struct { | |||
| EmitterName string `json:"emitterName,omitempty"` | |||
| EmitterMemBufLimit string `json:"emitterMemBufLimit,omitempty"` | |||
| EmitterStorageType string `json:"emitterStorageType,omitempty"` | |||
| // If true enable reloading via HTTP | |||
| HotReload *bool `json:"hotReload,omitempty"` | |||
There was a problem hiding this comment.
How to utilize this flag if enabled?
In which version it's added to fluentbit?
@wanjunlei pls take a look
There was a problem hiding this comment.
On how to use this flag can see https://docs.fluentbit.io/manual/administration/hot-reload
The version added to fluentbit should be v2.1 https://fluentbit.io/announcements/v2.1.0/
There was a problem hiding this comment.
We've already add --enable-hot-reload flag in fluentbit watcher:
https://github.com/fluent/fluent-operator/blob/master/cmd/fluent-watcher/fluentbit/main.go#L76C20-L76C39
And the config will be reload whenever there is a config change by sending SIGHUP signal:
https://github.com/fluent/fluent-operator/blob/master/cmd/fluent-watcher/fluentbit/main.go#L126
FluentBit doc says Fluent Bit supports the hot reloading feature when enabled via the configuration file or command line with -Y or --enable-hot-reload option. which means it's not necessary to add it to configfiles because the command line arg already works.
And fluent operator already supports HTTP settings https://github.com/fluent/fluent-operator/pull/1286/files#diff-7aad9bf9eeeaa93f3a61c5ea93ea5a30d57bcfa6307ec4ea188222fdf69f438cR99-R105
@wanjunlei what do you think? do we need to add the hot reload flag to configfiles?
There was a problem hiding this comment.
fluentbit already supports hot reload via the --enable-hot-reload flag, so there is no need to add Hot_Reload to theconfig file.
There was a problem hiding this comment.
There are several main reasons for adding this parameter:
- The reason for debugging. Because fluent-operator wraps the watcher layer, you can only restart the process by restarting the container or changing the configuration of the CR when debugging, restarting the container will be more troublesome and changing the configuration is not easy to locate the problem, so we need a way to send a signal to restart manually for hot loading.
- This parameter is one of the officially recommended ways to use this can also be harmonized with the official.
- Historical legacy issues, before the signal for hot loading in the watcher layer has been changed, due to the fluentbit version upgrade, so the signal to trigger hot loading has also been changed, and the way through the http is compatible with the lower version, but also for the sake of unity.
- Compared to the practice of adding parameters in the watcher layer, I think it is also a simple way to configure directly in the configuration file.
- Some users have suggested using multiple hot-loading methods before, so I think this is needed for some users, and this addition does not affect the original architecture.
There was a problem hiding this comment.
You have nicely explained this!
|
I think we're ready to release v3.1.0, are you able to trigger a new release? @wenchajun |
What this PR does / why we need it:
Hot reloading can be kicked via HTTP endpoints that are:
PUT /api/v2/reload
POST /api/v2/reload
Which issue(s) this PR fixes:
Fixes #
Does this PR introduced a user-facing change?
Additional documentation, usage docs, etc.: