Details
-
Improvement
-
Status: Closed
-
Normal
-
Resolution: Fixed
-
None
-
None
-
None
Description
I recently had some conversations with nick.walker and others about scaling PE to tens of thousands of nodes. One of the issues they ran into was that the disk got filled up by the YAML files containing the facts for nodes.
My belief was that these YAML files were necessary because of the way the indirector works for the catalog terminus; that it would save the facts out to YAML and then later on load them back from the YAML file to use in the compile.
However, in discussing this with eric.sorenson, henrik.lindberg and others, it seems like that is not actually the case. It seems like it's no longer required from a technical perspective, but was added back in because there were user workflows where people had built external tools that relied on scraping data from those YAML files.
In today's world, it's much more likely that people would want to use PuppetDB for those sorts of tasks, which means there is probably not much value in those files for most users.
We should consider having a setting to disable the creation of these files, or, if we already have some setting (node_cache_terminus?) that can do this, we should consider disabling them by default.
henrik.lindberg points out that this would also result in a performance improvement as these files can be rather large and we would no longer need to write them during catalog compiles.
Attachments
Issue Links
- relates to
-
PUP-7198 Puppet Server uses facts from the previous run (n-1) after agent switches masters
-
- Closed
-
-
SERVER-1819 Change node_cache_terminus default to nil (Server ticket)
-
- Closed
-
-
SERVER-2145 Disable server's fact cache by default
-
- Closed
-