/api/config.reload doesn't update the result of /api/config.dump

[mururu]

Check CONTRIBUTING guideline first and here is the list to help us investigate the problem.

• fluentd or td-agent version.

v1.0.x and v0.14.x

• How to confirm it.

$ cat rpc.conf
<system>
  rpc_endpoint 0.0.0.0:24444
</system>
$ bundle exec bin/fluentd -c rpc.conf
2018-01-15 15:17:35 +0900 [info]: parsing config file is succeeded path="rpc.conf"
2018-01-15 15:17:35 +0900 [info]: using configuration file: <ROOT>
  <system>
    rpc_endpoint "0.0.0.0:24444"
  </system>
</ROOT>
2018-01-15 15:17:35 +0900 [info]: starting fluentd-1.0.2 pid=37868 ruby="2.4.1"
2018-01-15 15:17:35 +0900 [info]: spawn command to main:  cmdline=["/Users/yito/.rbenv/versions/2.4.1/bin/ruby", "-Eascii-8bit:ascii-8bit", "-rbundler/setup", "bin/fluentd", "-c", "rpc.conf", "--under-supervisor"]
2018-01-15 15:17:35 +0900 [info]: gem 'fluentd' version '1.0.2'
2018-01-15 15:17:35 +0900 [info]: #0 starting fluentd worker pid=37901 ppid=37868 worker=0
2018-01-15 15:17:35 +0900 [info]: #0 fluentd worker is now running worker=0

--- call /api/config.dump ---

2018-01-15 15:18:03 +0900 [info]: dump in-memory config
2018-01-15 15:18:03 +0900 [info]: <ROOT>
  <system>
    rpc_endpoint "0.0.0.0:24444"
  </system>
</ROOT>

--- echo 'abc' >> rpc.conf ---
--- call /api/config.reload ---

2018-01-15 15:19:22 +0900 [info]: #0 fluentd worker is now stopping worker=0
2018-01-15 15:19:22 +0900 [info]: #0 shutting down fluentd worker worker=0
2018-01-15 15:19:22 +0900 [info]: Worker 0 finished unexpectedly with status 0
2018-01-15 15:19:23 +0900 [info]: gem 'fluentd' version '1.0.2'
2018-01-15 15:19:23 +0900 [warn]: parameter 'abc' in <ROOT>
  abc
  <system>
    rpc_endpoint "0.0.0.0:24444"
  </system>
</ROOT> is not used.
2018-01-15 15:19:23 +0900 [info]: #0 starting fluentd worker pid=37980 ppid=37868 worker=0
2018-01-15 15:19:23 +0900 [info]: #0 fluentd worker is now running worker=0

--- call /api/config.dump ---

2018-01-15 15:19:52 +0900 [info]: dump in-memory config
2018-01-15 15:19:52 +0900 [info]: <ROOT>
  <system>
    rpc_endpoint "0.0.0.0:24444"
  </system>
</ROOT>

[fujimotos]

I have investigated this issue and find that the current implementation
of /api/config.dump (and /api/config.getDump) is fairly broken.

The Root Problem

The root problem is that, the RPC server is running on a Server process
while only Worker processes know the actual configuration.

Since there is no IPC mechanism (except Unix signals) available for them
to communicate with each other, the RPC server, by design, never knows
the latest state of the configuration settings.

How can we fix this issue?

I can think of two ways to fix this, but I think neither is something
immediately feasible:

1. Add a new Unix signal handler to dump configuration settings
2. Implement a new IPC mechanism to allow communication between processes.

Probably, we'd better to disable these endpoints until they get fixed properly
(see also #1576). Provided that these endpoints are not documented,
degradation to users would be very small.

[github-actions]

This issue has been automatically marked as stale because it has been open 90 days with no activity. Remove stale label or comment or this issue will be closed in 30 days

[github-actions]

This issue was automatically closed because of stale in 30 days