1.0.0-beta.17 - #87

Merged

pathscale merged 72 commits into

Sep 3, 2026

Commits on Sep 2, 2026

refactor: extract the schema language into worktable_dsl
meh
committed
fdee8cc
Drop the redundant glob from the codegen shim
meh
committed
67c99f0
Read a schema as data, and write one back
meh
committed
43e4497
Say what changed between two schemas, and what it costs
meh
committed
b736579
Bake each table's declaration into the code it generates
meh
committed
43a24ba
Find the declarations in a source tree
meh
committed
0d0a6f7
Run the `pinned` example, rather than ignoring it
meh
committed
44bf36c
Express requirements as carets, and take ps-reclaim 0.1.1
meh
committed
8f775b0
Publish worktable_dsl, and hold the three crates in lockstep
meh
committed
b4d94ba
Accept a trailing comma after every block, and name the token that is not one
meh
committed
0414c6b
Let a schema be checked without expanding the macro
meh
committed
c77296b
Draw the line between the grammar and the model above it
meh
committed
0d2fe7f
Delete in batches, by key list or by span
meh
committed
ae005a5
Make the batch deletes async, like every other write
meh
committed
edd0fbf
Race writers across every backend, above the count the suite reached
meh
committed
fd6617a
Do not take a global mutex to learn the free list is empty
meh
committed
709fcb3
Keep the release-only guard out of clippy's constant-assertion lint
meh
committed
1f5c301
Race `insert` against `insert_many` on every backend
meh
committed
076dfe1
Verify every key width the macro advertises, and move the sweep to the benches
meh
committed
2eb4e7d
Free a run of links with one coalesce, not one per link
meh
committed
45a7b71
Stop rescanning the whole retirement queue on every sweep
meh
committed
689a3f8
Publish worktable and worktable_codegen to crates.vip

pathscale
authored and
meh
committed
27ff552
Point at the renamed backend repo

pathscale
authored and
meh
committed
c5d6f1d
Publish worktable_dsl to crates.vip too, and before codegen
meh
committed
511cf88
Propose a way to observe when a write is actually durable
meh
committed
5f075a2
Let a delete be the bit flip, and leave the cleanup to whoever needs it
meh
committed
5b5ce8e
Bound the torn-shutdown writer's range
meh
committed
679802a
Race readers against storage reuse, not just against inserts
meh
committed
453ebcc
Record the branches that existed only on one machine
meh
committed
0025e95
Report an index over a missing column, rather than panicking on it
meh
committed
fe9f95e
Close two review blockers: a reclamation race and a parser panic
meh
committed
c942c77
Make persisted bulk deletes actually reach the disk
meh
committed
e2b6294
Close the remaining review findings in deletion, the free list and the DSL
meh
committed
e7a7beb
Let upsert insert without letting go of the row it is deciding about
meh
committed
88a1d1b
Make insert async, like every other write on this table
meh
committed
aab5f4c

Commits on Sep 3, 2026

Make the embedded schema readable by the crate that embeds it
meh
committed
3801bc2
Plan a primary key generator change as a change
meh
committed
6ee1671
Add wt-dsl, so the schema language has a command line
meh
committed
e3bc3c0
Fold in the DSL emitter binaries, and take the correction with them
meh
committed
b79353f
Check the primary key's backend and generator, which check never did
meh
committed
ff2df9c
Release worktable_dsl as 1.0.0-beta.18, and localise the vacuum failure
meh
committed
71caaaa
Do not reclaim a page a row was inserted onto while it was being drained
meh
committed
c4fc1bf
Write down what the vacuum bugs have in common
meh
committed
5547916
Say which of the three backends the vacuum note is talking about
meh
committed
1cf8285
Assert what vacuum must not break, on all three backends
meh
committed
9d571d3
Plan vacuum from free ranges, which the registry already maintains
meh
committed
ddd7ba7
Give arctic and congee an `iter`, so the backends are interchangeable
meh
committed
a9eecc0
Make vacuum reactive and stop it holding the registry for a whole sweep
meh
committed
707d29b
Arctic takes signed keys
meh
committed
97d30a7
Ignore the persisted-vacuum reproduction, and say what it reproduces
meh
committed
1454448
Recover batch collection when event order and operation order disagree
meh
committed
67fa8c5
Pin the publish workflow to a commit sha
meh
committed
860c758
Two doc comments describing code that had already changed
meh
committed
2af637b
Separate widening the batch search from giving up on it
meh
committed
c4de85f
WT-6: a bulk delete no longer holds every stripe for its whole duration
meh
committed
2fc5be9
A mutation sweep and an adversarial corpus over the DSL check
meh
committed
ffca5b3
Retract the tail claim: a clean re-run does not reproduce it
meh
committed
0fa0522
WT-8: fail CI on a duplicated WorkTablesIndex or data_bucket
meh
committed
ac51393
Vacuum stress on the stack that will actually ship
meh
committed
761e5d5
Stand down exponentially, which is what the 60% case needed
meh
committed
fbedd4b
Let a delete burst settle before the woken sweep runs
meh
committed
20414b3
Revert "Stand down exponentially, which is what the 60% case needed"
meh
committed
9a92003
Remove the polling interval from the vacuum manager's config
meh
committed
760f829
Say what the stall diagnostic now knows
meh
committed
587b957
Reapply "Stand down exponentially, which is what the 60% case needed"
meh
committed
82c686c
Revert "Reapply "Stand down exponentially, which is what the 60% case needed""
meh
committed
6144ed0
Record what the vacuum manager actually did
meh
committed
41b63c1
Expose the page counts that verify a vacuum
meh
committed
bb5beef
Sweep until the table is clean, not once per wake
meh
committed
9ae13a6
Ask the table if it is busy, instead of guessing from space demand
meh
committed
decdd9f
Finish the beta17 mutation and DSL validation fixes
meh
committed
af9f26c
Record the beta17 validation evidence
meh
committed
2f3b01a