Windows: handle.exe output did not contain a line beginning with 'File', so parsing of its output fails

[rgl]

Describe the bug

when running rabbitmq 4.0.2 on windows 2019, the rabbitmq logs contain these warnings:

2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0> handle.exe output did not contain a line beginning with 'File': ["Nthandle v5.0 - Handle viewer",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "Copyright (C) 1997-2022 Mark Russinovich",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "Sysinternals - www.sysinternals.com",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "Handle type summary:",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 2",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 20",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 87",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 1",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 6",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 6",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 2",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 1",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 3",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 6",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 8",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 53",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 1",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 14",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 5",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "<Unknown type>  : 86",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "Process         : 4",
2024-09-23 17:44:21.550000+01:00 [warning] <0.1325.0>                                                                  "Total handles: 305"]

this also happens with handle.exe v4.22.

Reproduction steps

1. Install latest handle.exe (at time of writing this was 5.0) from https://learn.microsoft.com/en-us/sysinternals/downloads/handle.
2. On Windows 2019, install RabbitMQ 4.0.2.
3. Start RabbitMQ
4. Wait for a few minutes.
5. Consult the RabbitMQ logs, and observe the above warning.

FWIW, this warning comes from https://github.com/rabbitmq/rabbitmq-server/blob/v4.0.2/deps/rabbitmq_management_agent/src/rabbit_mgmt_external_stats.erl#L122.

Expected behavior

No warnings in the logs.

Additional context

No response

[lukebakken]

Yep, this happens on my Windows 11 development laptop as well. It's because handle.exe does not return the expected output from time-to-time. I've contacted @markrussinovich about it but he's obviously busy with way more important items than SysInternals or handle.exe

Team RabbitMQ has put together a replacement for handle.exe (https://github.com/rabbitmq/fshc), maybe it's time to integrate that with RabbitMQ.

Thoughts @michaelklishin @the-mikedavis ?

[michaelklishin]

Yes, this is the same conclusion I have arrived to. Sometimes the output is simply not stable and we don't know why.

Exactly not to depend on handle.exe for our small needs, we have developed https://github.com/rabbitmq/fshc but then 4.0 priorities have gotten in the way.

Once fshc is built and published as a release, we can adapt RabbitMQ to support it as an alternative to handle.exe.

[rgl]

How should this be tracked? Can you convert this github discussion to an issue?

[michaelklishin]

@rgl we will not convert this into an issue because it's not something our team is responsible for or can fix. handle.exe is a piece of closed source abandonware.

The consequence of this behaviour comes down to one metric not being available. The kind that should be monitored at the OS level anyway.

Our tiny replacement for our use of handle.exe is open source and can be watched like any other repo. You can help integrate it into RabbitMQ, this is open source software after all.

[rgl]

Sorry, but I don't quite get your take on this. Without an issue, how is this project supposed to track pending changes?

IMHO, since the use of handle.exe is integrated in this project source code, it's this project responsibility to handle it.

In hindsight, if using handle.exe no longer makes sense, it should either be replaced or removed.

[michaelklishin]

@rgl handle.exe is an entirely optional "feature" (dependency) specific to Windows. We never promised any "pending changes", fshc is a hobby project of a couple of engineers on our team in their free time.

Because it's really tiring to "be responsible" a behavior of a closed source, 20+ year old tool that is 1000% Windows-specific (file handle metrics is a solved problem on other OSes we support) and up until recently, did not have any open source alternatives known to me.

We have developed a small open source alternative, which though this is by no means a priority for our team. If someone cares enough about "tracking" fshc integration, the best way to do that is to help integrate it into rabbit_mgmt_external_stats, first alongside handle.exe and eventually as the only supported option.

Or just run RabbitMQ on Linux, where this metric is available without any additional tools.

[lukebakken]

#12369