Purging single file using url through API says OK but still stale


Been successfully purging content by url using the CloudFlare API for over two years.

However since 1am October 5th I can see issues from our logs of publishing times (which includes a purge to update files).

After ruling out varnish caching I can confirm that content that is refreshed and purged from CloudFlare (response logged as OK) is still stale in CloudFlare.

This is an example url that served stale content
that had a cf-ray

I do programmatic checks against
Which i confirmed was 2019-10-22T13:47:01.96Z in the url serving CF by directly calling the Varnish server for their content.

Obviously CF can take some time to purge, however, I retry the check for purge for 60 seconds with subsequent purge commands run. It sometimes works fine but it seems that almost 50% of the time purging fails. Can purging not happen if I get an OK? Is there a way to find out?

I have logs that show this started almost exactly at the beginning of Oct 5th. I have looked every else other than CF as it is usually so reliable and indeed usually only took a couple of seconds to purge, possibly because our content is probably mostly cached in Dublin.