* ⚡ Improve efficiency on objects-gc tasks
Replacing db/cursor with db/plan
* 🐛 Use correct order of subprocs on objects-gc
Mainly affects the file deletion
* 📎 Increase chunk-size to 100 on objects-gc
This upgrade also includes complete elimination of use spec
from the backend codebase, completing the long running migration
to fully use malli for validation and decoding.
The main issue was the long running gc operation that
affects storage objects with deduplication. The long running
transacion ends locking some storage object rows which collaterally
made operations like import-binfile become blocked indefinitelly
because of the same rows (because of deduplication).
The solution used in this commit is split operations on small
chunks so we no longer use long running transactions that holds
too many locks. With this approach we will make a window to work
concurrently all operarate the distinct operations that requires
locks on the same rows.
Relevant changes:
- Add the ability to create migration in both directions, defaulting
to identity if not provided
- Move the version attribute to file table column for to make it more
accessible (previously it was on data blob)
- Reduce db update operations on file-update rpc method
Previously the file processing was implemented 3 times using similar
approaches bug each own with its own bugs. This PR unifies the
loging to a single implementation used by the 3 operations.
The main objective is prevent deletion of objects that can leave
unreachable orphan objects which we are unable to correctly track.
Additionally, this commit includes:
1. Properly implement safe cascade deletion of all participating
tables on soft deletion in the objects-gc task;
2. Make the file thumbnail related tables also participate in the
touch/refcount mechanism applyign to the same safety checks;
3. Add helper for db query lazy iteration using PostgreSQL support
for server side cursors;
4. Fix efficiency issues on gc related task using server side
cursors instead of custom chunked iteration for processing data.
The problem resided when a large chunk of rows that has identical
value on the deleted_at column and the chunk size is small (the
default); when the custom chunked iteration only reads a first N
items and skip the rest of the set to the next run.
This has caused many objects to remain pending to be eliminated,
taking up space for longer than expected. The server side cursor
based iteration does not has this problem and iterates correctly
over all objects.
5. Fix refcount issues on font variant deletion RPC methods
Mainly move all pointer-map related helpers from app.rpc.commands.files
to the the app.features.fdata namespace and normalizes codestile around
feature handling on all affected code.
This commit also comes with several features related bugifxes on the
components-v2 migration code:
- properly migrate legacy feature names on apply components-v2 migration
- start using new fdata feature related functions
- prevent generation of a ephimeral pointer on each graphic migration
operation; on large files this caused a very noticiable overhead of
creating a big number of completly unused pointer maps
- do persistence after validation and not before
