Varnish/ATS are processing on average 1.8k PURGE requests per second on the text cluster. To put this into perspective, ATS handles more PURGE requests than GETs per server on any DC and it's not uncommon to see PURGE requests matching and exceeding the volume of GET requests being handled by varnish in a busy DC like eqiad
An initial 60 seconds capture on one of the test host showed the following top 10 hostnames:
```
sort purge.log |uniq -c | sort -rn | head -10
52670 commons.wikimedia.org
34983 commons.m.wikimedia.org
34533 tr.wikipedia.org
25459 en.wikipedia.org
9296 en.m.wikipedia.org
1710 de.wikipedia.org
1584 www.wikidata.org
1446 m.wikidata.org
990 uk.wikipedia.org
825 fr.wikipedia.org```
A following 300 seconds capture on the same host shows
```
$ sort purge.log |uniq -c | sort -rn | head -10
91377 en.wikipedia.org
54596 he.wikipedia.org
15094 en.m.wikipedia.org
10590 www.wikidata.org
10590 m.wikidata.org
9060 he.m.wikipedia.org
8972 commons.wikimedia.org
8434 fr.wikipedia.org
8016 pl.wikipedia.org
7230 de.wikipedia.org
```
non-PURGE requests VS PURGE requests hitting cp3050 during the last 30 days:
{F35528462}