secantus-mdb 0.5.3-beta.182 gives time-series collections their buckets and makes _id respect the collation

Summary: secantus-mdb 0.5.3-beta.182 shows a time-series collection's buckets, makes _id unique by the collection's default collation, stops a failed $out from destroying its target, and adds convertToCapped, cloneCollectionAsCapped, compact and reIndex.

Two things in this release could leave a test passing on wrong data.

• A failed $out emptied its target. The target was dropped before the new documents went in. If one of them was refused, by a duplicate _id or by a unique index the output broke, the target was left empty or half-written, and a duplicate _id in the output was dropped without an error. $out now builds the replacement beside the target and swaps it in, as mongod does, so a refused output leaves the target as it was.
• _id was not unique by the collection's default collation. With a default of {locale: "en", strength: 2}, mongod refuses {_id: "A"} once {_id: "a"} is there. secantusd-rs accepted it, on insert and on upsert. It refuses it now, with mongod's error. A collection created with a default collation by an earlier release is not protected for the documents already in it.

$merge had the same blind spot: it matched by a simple comparison whatever the target's collation, and so inserted a second spelling. It matches under the target's collation now and refuses to change the _id it matched (66), as mongod does.

Time-series collections. mongod stores a time-series collection as bucket documents in system.buckets.<name> and presents the collection as a view over them. This server stores the measurements, in the format the Python server shares, and until now had no buckets to show. It derives them now, whenever something asks:

• system.buckets.<name> and system.views can be read, counted, listed and used as the base of a view. A bucket has mongod's shape: control.min and control.max, meta, and data columns that mongod 8.2.11 unpacks to the same measurements.
• A scan of the collection returns its measurements bucket by bucket, each with the time field first and the meta field second, as mongod does.
• validate, collStats and $collStats report buckets, and listIndexes on the buckets shows each index as it lies on them.
• A document that is not a bucket is refused by the bucket schema (121).

The comparison with mongod 8.2.11 covers 187 time-series scenarios; the previous release differed on 24, this one on 6. The buckets are a derivation, not storage: a bucket cannot be written directly, and mongod places a batch of measurements as a whole where this server places them one by one, so an insert batch that runs backwards in time can land in more buckets here. The release notes in tasks/backlog.md list the rest.

Four commands that answered no such command. convertToCapped and cloneCollectionAsCapped copy a collection into a capped one, keeping its options and the newest documents the cap allows, with mongod's checks and errors. compact and reIndex answer as mongod does: refused on a replica-set member, which this server is unless started with --standalone, and bytesFreed: 0 / the index list otherwise.

Smaller changes: a duplicate key on a clustered collection is worded as mongod words it, with foundValue; validate reports repairMode, the collection's uuid and each index's spec; $out to a capped collection is refused (17152).

We ran the pymongo, Go and Node driver suites against this code. Each gave the same result as the release before.

This is a server for tests. It is single-node and it is a beta.

cargo install secantus-mdb --version 0.5.3-beta.182 builds it from crates.io. Binaries for Linux x86_64, macOS arm64 and Windows x86_64 are on the release.

Rust MongoDB server · secantus-mdb on crates.io · MongoDB binaries