Jenkins triggers SCM polling against last successful build on restart

[AndresF]

Hi all,

I've come across this strange behaviour:

I have serveral projects configured for SCM polling (Subversion). They work well, triggering the build when changes happen on the remote repository. Even when the last build is FAILURE, it uses it's SCM revision as a baseline for comparing revisions against the remote repository.

However, at Jenkins restart, polling is triggered, but apparently it compares against the LAST SUCCESSFUL build of the project. Also, Jenkins keeps always the last successful build of every project, regardless of how many builds it's configured to keep.

In projects that fail often, this can lead to comparing to an old revision at restart, triggering the build.

Is this really a lack of consistency in the polling mechanism or am I missing something?