expiration not working as expected

[rabernat]

Here's a reproducer

import icechunk as ic
import zarr
from datetime import datetime, timezone

storage = ic.local_filesystem_storage("./expiration.icechunk")
repo = ic.Repository.create(storage)
session = repo.writable_session("main")
array = zarr.create_array(store=session.store, shape=100, chunks=10, dtype='i4')
session.commit("created array")
session = repo.writable_session("main")
array = zarr.open_array(session.store)
array[:] = 1
session.commit("wrote_data")
print(repo.ancestry(branch="main"))

At this point I have 3 commits

[SnapshotInfo(id="4KHFWPG444PRDN9JJYC0", parent_id="1B48F0SVAPJH6FRQHRG0", written_at=datetime.datetime(2025,2,14,3,28,55,897712, tzinfo=datetime.timezone.utc), message="wrote_data..."),
 SnapshotInfo(id="1B48F0SVAPJH6FRQHRG0", parent_id="G4WNZE9XN03X1E26JVB0", written_at=datetime.datetime(2025,2,14,3,28,55,890476, tzinfo=datetime.timezone.utc), message="created ar..."),
 SnapshotInfo(id="G4WNZE9XN03X1E26JVB0", parent_id=None, written_at=datetime.datetime(2025,2,14,3,28,55,886074, tzinfo=datetime.timezone.utc), message="Repository...")]

I should be able to expire two of them.

repo.expire_snapshots(datetime.now(timezone.utc))
# -> set()

But this returns an empty set.

[rabernat]

Ok so I see that the problem is that I gave it a date after the latest snapshot.

If I do

repo.expire_snapshots(anc[0].written_at)
# -> {'1B48F0SVAPJH6FRQHRG0'}

it expires the middle snapshot.

---

I guess I was expecting that it would automatically keep the snapshot pointed by main and expire everything else

---

Any reason why the first snapshot is not expired?

[paraseba]

I think that would be reasonable behavior. I'll need to think if it impacts the algorithm in any way

[rabernat]

It's worth noting that expiring everything but the latest snapshot is probably very dangerous, because there could be other sessions active on the recent snapshots.