Use a less coupled vault architecture
ClosedPublic
Actions

Authored by seirl on Nov 7 2017, 6:24 PM.

Details

Reviewers

ardumont

Summary

Cooking tasks can now be started on different machines/configurations
than the backend.

Repository

rDVAU Software Heritage Vault

Lint

No Linters Available

Unit

No Unit Test Coverage

Build Status

Buildable 1090
Build 1433: arc lint + arc unit

Except for the cursor detail which i don't get, this seems fine.

swh/vault/cookers/revision_gitfast.py
45–46	Why?
50	I'm not sure to understand why we see cursor mentioned here. The cursor is an implementation detail of the storage. Plus, like you mentioned, there are cases where we cannot exploit this. And also, if we do like for the loaders (and probably listers), we will mostly use the remote storage.

swh/vault/cookers/revision_gitfast.py
45–46	Because using a single transaction for the whole query should be faster, shouldn't it?

swh/vault/cookers/revision_gitfast.py
45–46	Probably. How can we know though without testing? And maybe you did for huge repositories, i don't know :) In any case, i believe the the comment should say so and not let the reader guessing why. `To be faster, we want a single transaction...` Otherwise, if it is that obvious, then the comment is not needed.