perf(grpc): batch ListMessages placements; fix JMAP parse folder #418

2026-08-15T00:44:23Z

KofTwentyTwo commented

2026-08-15 00:44:23 +00:00

(Migrated from github.com)

Two related pieces of placement cleanup.

1. `ListMessages` was N+1 on placements

fill_placements called placements_of once per message, so a page of fifty
cost fifty-one queries against the store.

Adds DatabaseEngine::placements_of_batch, which reads a whole set of message
keys and returns them grouped by key, and uses it from the listing path.

Chunked at the existing MAX_KEYS_PER_IN_LIST, mirroring
existing_placements, so the SQLite bind-parameter ceiling stays unreachable
from outside the store. Duplicate keys are collapsed before chunking, so no
key's placements can be gathered from two chunks and appended twice.
Ordering preserved. Each key's placements come back ordered by folder id
ascending, exactly as placements_of orders them. primary_placement takes
the first, and the proto documents Message.folder_id as the lowest-sorting
folder id, so this is contract, not incidental.

New test placements_of_batch_matches_per_message_placements_of_including_order
proves the batched read equals the per-message read element for element for a
multi-message, multi-placement fixture, asserts folder-id order explicitly,
and covers the repeated-key and unrequested-key cases.

Error handling changes shape: a failed batch now leaves every set on the page
empty and logs once, where before a failure was per message. Both degrade to a
page without placements rather than a failed listing, and the field is additive.

2. `parse_email_get_response` no longer hardcodes `folder_id = "inbox"`

The production path was already corrected in #414 (placements are labelled with
the mailbox actually being synced), but the public helper kept its fixed-folder
signature. It now takes the folder id it is parsing for, and the private
parse_email_get_response_in_folder variant the sync loop used to work around
the fixed label is gone — there is one function again. Callers say which mailbox
they mean.

Closes #339

Gate

cargo fmt --all -- --check, cargo check-all, cargo test-all all green.
RUST_TEST_THREADS=2 cargo test-all run three times: 0 / 0 / 0 FAILED lines.

Two related pieces of placement cleanup. ## 1. `ListMessages` was N+1 on placements `fill_placements` called `placements_of` once per message, so a page of fifty cost fifty-one queries against the store. Adds `DatabaseEngine::placements_of_batch`, which reads a whole set of message keys and returns them grouped by key, and uses it from the listing path. - **Chunked** at the existing `MAX_KEYS_PER_IN_LIST`, mirroring `existing_placements`, so the SQLite bind-parameter ceiling stays unreachable from outside the store. Duplicate keys are collapsed before chunking, so no key's placements can be gathered from two chunks and appended twice. - **Ordering preserved.** Each key's placements come back ordered by folder id ascending, exactly as `placements_of` orders them. `primary_placement` takes the first, and the proto documents `Message.folder_id` as the lowest-sorting folder id, so this is contract, not incidental. New test `placements_of_batch_matches_per_message_placements_of_including_order` proves the batched read equals the per-message read element for element for a multi-message, multi-placement fixture, asserts folder-id order explicitly, and covers the repeated-key and unrequested-key cases. Error handling changes shape: a failed batch now leaves every set on the page empty and logs once, where before a failure was per message. Both degrade to a page without placements rather than a failed listing, and the field is additive. ## 2. `parse_email_get_response` no longer hardcodes `folder_id = "inbox"` The production path was already corrected in #414 (placements are labelled with the mailbox actually being synced), but the public helper kept its fixed-folder signature. It now takes the folder id it is parsing for, and the private `parse_email_get_response_in_folder` variant the sync loop used to work around the fixed label is gone — there is one function again. Callers say which mailbox they mean. Closes #339 ## Gate `cargo fmt --all -- --check`, `cargo check-all`, `cargo test-all` all green. `RUST_TEST_THREADS=2 cargo test-all` run three times: 0 / 0 / 0 `FAILED` lines.

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: KofTwentyTwo/nuncio#418

perf(grpc): batch ListMessages placements; fix JMAP parse folder #418

1. ListMessages was N+1 on placements

2. parse_email_get_response no longer hardcodes folder_id = "inbox"

Gate

1. `ListMessages` was N+1 on placements

2. `parse_email_get_response` no longer hardcodes `folder_id = "inbox"`