File

mod_log_slow_events/README.md @ 6057:cc665f343690

mod_firewall: SUBSCRIBED: Flip subscription check to match documentation The documentation claims that this condition checks whether the recipient is subscribed to the sender. However, it was using the wrong method, and actually checking whether the sender was subscribed to the recipient. A quick poll of folk suggested that the documentation's approach is the right one, so this should fix the code to match the documentation. This should also fix the bundled anti-spam rules from blocking presence from JIDs that you subscribe do (but don't have a mutual subscription with).
author Matthew Wild <mwild1@gmail.com>
date Fri, 22 Nov 2024 13:50:48 +0000
parent 6003:fe081789f7b5
line wrap: on
line source

---
labels:
- 'Stage-Stable'
summary: Log warning when event handlers take too long
...

Introduction
============

Most activities in Prosody take place within our built-in events framework, for
example stanza processing and HTTP request handling, authentication, etc.

Modules are able to execute code when an event occurs, and they should return
as quickly as possible. Poor performance (e.g. slow or laggy server) can be caused
by event handlers that are slow to respond.

This module is able to monitor how long each event takes to be processed, and
logs a warning if an event takes above a certain amount of time, including
providing any details about the event such as the user or stanza that triggered it.

The aim is to help debug why a server may not be as responsive as it should be,
and ultimately which module is to blame for that.

Configuration
======================

There is a single configuration option:

```
   -- Set the number of seconds an event may take before
   -- logging a warning (fractional values are ok)
   log_slow_events_threshold = 0.5
```

Metrics
=======

In addition to the log messages, a new 'slow_events' metric will be exported to
your configured stats backend (if any).

Compatibility
-------------

  ------- --------------
  trunk   Works
  0.10    Works
  0.9     Doesn't work
  0.8     Doesn't work
  ------- --------------