With this proposal, I just remembered something.
Apache SkyWalking provided a mechanism which called: Thread Profiling
<https://skywalking.apache.org/docs/skywalking-python/next/en/profiling/profiling/#how-does-threading-profiler-the-default-mode-work>
The core of the mechanism is start a scheduler to periodic take the thread
snapshots(stackTrace), say, every 1/5/10 ms take a snapshot.
Then, group these snapshots.
Say, I’ll take 1 snapshot every 5ms, and take 3 snapshots:
1:
```
method call C
method call B
method call A
```
2.
```
method call C
method call B
method call A
```
3.
```
method call D
method call B
method call A
```
Then, we can group these snapshots to:
```
method call C -10ms method call D -5ms
method call B
method call A
```
We can apply the feature to CLI Tools or Admin API. And we can even specific a
thread name prefix to profile these threads which name starts with the prefix.
In the past, we use `jstack <pid>` to take threads dump, even need to take
threads dump many times and manually compare these threads dumps to locate
where blocks. With the feature, we can locate where blocks easier.
This proposal tells us there are some threads blocks, and the above feature
tells us where blocks. It can be an effective supplement to the proposal
Thank,
Tao Jiuming
> 2022年12月19日 20:17，adobewjl <wjl_is_...@163.com <mailto:wjl_is_...@163.com>>
> 写道：
>
> Hello pulsar community,
> I've opened `PIP-232: Introduce thread monitor to check if thread is blocked
> for long time.` to discuss.
> For more details, please read the PIP at
> https://github.com/apache/pulsar/issues/18985
> <https://github.com/apache/pulsar/issues/18985>
> I'm looking forward to hearing what you think.
> Also the demo PR link at https://github.com/apache/pulsar/pull/18958
> <https://github.com/apache/pulsar/pull/18958>
