vdbergh
vdbergh
Serializing of a run with 500 tasks only takes a couple of ms, but I am worried the middleware will wreak havoc with the timings. It would be nice to...
I was more thinking of something basic like shared memory.
Instead of returning the serialized run in the body, the primary instance could write it to shared memory where the secondary instance would find it. But it is not so...
Another option is a named pipe. The primary instance creates it, writes the serialized run to it and communicates the name to the secondary instance. The secondary instance reads it,...
Of course we may just as well use a real temporary file for this then.
Well #1983 is something else. This PR is because in #1985 and #1992 the run read from the database may be missing some tasks compared to what's in the cache....
``` return self.runs.find_one({"_id": ObjectId(r_id)}) ``` I don't understand. This still seems to return the run from the db, which is not what you want...
> But datetime and id must be converted before the dump. One can use bson instead of json. bson doesn't have this issue.
@ppigazzini Your code seems to be much more ambitious than what I am trying to do in this PR. - In this PR I want to make it possible for...
In the latest push I skip the base64 encoding. This makes the payload smaller. EDIT: I should perhaps point out that runs compress extremely well (obviously because there is so...