diff options
Diffstat (limited to 'doc/v7compatibility.html')
| -rw-r--r-- | doc/v7compatibility.html | 37 |
1 files changed, 37 insertions, 0 deletions
diff --git a/doc/v7compatibility.html b/doc/v7compatibility.html index 8834cd54..24c17d72 100644 --- a/doc/v7compatibility.html +++ b/doc/v7compatibility.html @@ -51,6 +51,43 @@ errors and error messages. Starting with 7.2.1, this has been reduced to 10 successive failures. This still gives the plugin a chance to recover. In extreme cases, a plugin may now enter suspend mode where it previously did not do so. In practice, we do NOT expect that. +<h1>Notes for the 7.3/7.4 branch</h1> +<h2>"last message repeated n times" Processing</h2> +<p>This processing has been optimized and moved to the input side. This results +in usually far better performance and also de-couples different sources +from the same +processing. It is now also integrated in to the more generic rate-limiting +processing. +<h3>User-Noticable Changes</h3> +The code works almost as before, with two exceptions: +<ul> +<li>The supression amount can be different, as the new algorithm + precisely check's a single source, and while that source is being + read. The previous algorithm worked on a set of mixed messages + from multiple sources. +<li>The previous algorithm wrote a "last message repeated n times" message + at least every 60 seconds. For performance reasons, we do no longer do + this but write this message only when a new message arrives or rsyslog + is shut down. +</ul> +<p>Note that the new algorithms needs support from input modules. If old +modules which do not have the necessary support are used, duplicate +messages will most probably not be detected. Upgrading the module code is +simple, and all rsyslog-provided plugins support the new method, so this +should not be a real problem (crafting a solution would result in rather +complex code - for a case that most probably would never happen). +<h3>Performance Implications</h3> +<p>In general, the new method enables far faster output procesing. However, it +needs to be noted that the "last message repeated n" processing needs parsed +messages in order to detect duplicated. Consequently, if it is enabled the +parser step cannot be deferred to the main queue processing thread and +thus must be done during input processing. The changes workload distribution +and may have (good or bad) effect on the overall performance. If you have +a very high performance installation, it is suggested to check the performance +profile before deploying the new version. Note: for high-performance +environments it is highly recommended NOT to use "last message repeated n times" +processing but rather the other (more efficient) rate-limiting methods. These +also do NOT require the parsing step to be done during input processing. <p><font size="2">This documentation is part of the <a href="http://www.rsyslog.com/">rsyslog</a> project.<br> |