Affects Version/s: None
Fix Version/s: None
In the eventual/_queue folder, there is no segregation between delete's and add's.
This is in case of no-NFS "cluster" configurations.
However, in past NFS configurations, inside the eventual there was always separation between pre/add/delete.
This separation is important, because in situations where there are lots of "delete" in the eventual, this can impact the performance of uploads/downloads, which was not the case when there was separation.
Removed 5,000,000 artifacts from Artifactory, then run GC.
GC caused in 5,000,000 files inside the eventual/_queue folder.
Due to this, now all threads, even just upload/download is slow due to:
If we had separation as in the past, this would not happen.
Alternatively, if we can create sub-folders in the eventual/_queue, with checksum prefixes such as in filestore, this may help too.