This might be actual job configurations but no idea how to debug further, thus tagging Continuous-Integration-Config for the time being.
According to https://phabricator.wikimedia.org/diffusion/pulllog/ , integration-agent-docker-*.integration.eqiad1.wikimedia.cloud machines pull certain repos from Diffusion about 90 times within 60 minutes which seems excessive.
From a quick look, this seems to only be the case for 3 and a half repos.
Data from the last 30 days, not filtered on IPs (ofc there's also "normal" pulling from 3rd party or internal codesearch.eqiad1.wikimedia.cloud):
mysql:phstats@m3-slave.eqiad.wmnet [phabricator_maniphest]> SELECT COUNT(r.name) AS pulls, r.name FROM phabricator_repository.repository_pullevent rp INNER JOIN phabricator_repository.repository r ON r.phid = rp.repositoryPHID GROUP BY r.name ORDER BY pulls DESC LIMIT 9; +-------+-----------------------------------+ | pulls | name | +-------+-----------------------------------+ | 47199 | DataValues Javascript | | 47196 | Wikibase Data Model JavaScript | | 47195 | Wikibase Serialization JavaScript | | 12138 | new lexeme special page | | 3836 | netbox-exported-dns | | 1407 | wikibase-termbox | | 322 | MediaWiki | | 293 | skin-Vector | | 122 | extension-HeaderTabs | +-------+-----------------------------------+ 9 rows in set (0.249 sec)