Closed jimmymathews closed 1 year ago
As of 0536244d3938bc2f669bf06278a90646ddfc0ae6 (branch issue219) , the per-specimen during-processing serialization is enabled in spt ondemand cache-expressions-data-array
(for the feature matrix part, not the centroids/positional info).
As of 4364ea99a234e2fe8a9e3da63a2887e238585e83 (branch issue219), enabled for the centroids as well.
I think the memory adjustment has solved this issue. The cache creation script completes and proceeds.
However, it seems there is a problem with the volume mounting. Files created by one container (counts) are only sometimes available from other containers or at different times.
This leads to the problem that the other containers (proximity, squidpy) never start up, because they are waiting to see the cache files created.
This latter problem is really related to spt-dashboard #63
The primary issue described here is resolved, I think.
Currently there are application initialization steps (indexing, cache creation) which are still unnecessarily resource-intensive and cannot complete correctly in some environments. For this issue, improve the centroid cache creation and the expression matrix cache creation to do per-sample serialization-to-file during processing, in order to vastly decrease RAM usage requirements.