@procedures/store 0.39.0 → 0.41.0

package/CHANGELOG.md

@@ -1,5 +1,32 @@
 # @procedures/store
 
+## 0.41.0
+
+### Patch Changes
+
+- Updated dependencies [37deb07]
+- Updated dependencies [fc52431]
+  - @procedures/client@0.41.0
+  - @procedures/collections@0.41.0
+
+## 0.40.0
+
+### Minor Changes
+
+- 877b6aa: `HybridStorage`: the default `isUnavailable` gives true for a `TIMEOUT` (roadmap O68). Before, a `write-through` write that timed out was undone and threw the `TIMEOUT`, though the remote can have applied it. At this time, the write stays in the local copy, goes in the queue, and throws `PendingWriteError`. The drain sends it again. The collection writes set or delete a value by id, so a second send gives the same item. With `versionOf`, the push sends the base version, and an earlier send that reached the remote is not a conflict. The push then sends the same value one more time, with the current version as its precondition. That write changes no value and no version, but a subscriber of the remote can get a second change event for the item.
+  
+  Migration steps:
+  
+  - The site: no step. It does not catch a `TIMEOUT` of a `HybridStorage` write.
+  - MiniMongo: no step, unless code catches a `TIMEOUT` of a `write-through` write to show an undo. That code gets `PendingWriteError`, with the `TIMEOUT` as its `cause`.
+  - Graph: the own `isUnavailable` that keeps `TIMEOUT` in the queue is not necessary at this time. Remove it, or keep it: the result is the same.
+  - jarvis: no step. To keep the old behavior, pass `isUnavailable: (error) => error?.retryable === true || error?.code === undefined`.
+
+### Patch Changes
+
+- @procedures/client@0.40.0
+  - @procedures/collections@0.40.0
+
 ## 0.39.0
 
 ### Patch Changes

package/README.md

@@ -181,6 +181,7 @@ await users.setMode("write-through"); // The switch sends the queue, then syncs
 
 - In the modes `local-only` and `write-back`, each write goes in the queue.
 - In the mode `write-through`, a write that the remote cannot take goes in the queue, and the write throws `PendingWriteError`.
+- By default, a write that the remote cannot take is an error with `retryable: true`, an error without a `code`, or a `TIMEOUT`. The option `isUnavailable` changes it.
 - A full queue refuses the next write with `QueueFullError`. The limit is `maxOfflineOps`, 1000 by default.
 - A push keeps a write that the remote cannot take at this time (`isUnavailable` gives true). This write and the later writes stay in the queue.
 - A queued write that the remote refuses leaves the queue, and the push goes on with the next write.