After the WiredTiger import in MongoDB (0c6ba8d8be02dd34a46c3e9533971f1739b6ad8e), WT's cache while running as an in-memory storage engine does not reconcile memory promptly causing the test to fail with a `WT_CACHE_FULL` error.
A build failure sprung up as a consequence of this, where it requires approximately 5-6GB of cache to pass successfully, as opposed to ~3GB prior to the WT import.
I'm attaching diagnostic data I collected of the test (agg_sort_external.js) from before and after the WT import, as well as some graphs that could be useful to understanding the problem.
From my observations, after the import, the WT cache fill ratio is climbing steadily after a certain point and the tracked bytes belonging to leaf pages in the cache climbs towards the 4GB cache limit, causing the test to fail.
Prior to the import, the WT cache fill ratio would oscillate around 50%, as well as the tracked bytes belonging to leaf pages in the cache at around 2GB.
`metrics.passing` was taken from `970e0c02de8c5922efcab74cfee258f447f94741` and `metrics.failing` was taken from `83336cb56b269195110253918d226cbba4377a03` on the MongoDB repo.