feat: room server — reconnect sync fix, pubkey alias system, name resolution endpoint#220
Closed
tjdownes wants to merge 7 commits intorightup:devfrom
Closed
feat: room server — reconnect sync fix, pubkey alias system, name resolution endpoint#220tjdownes wants to merge 7 commits intorightup:devfrom
tjdownes wants to merge 7 commits intorightup:devfrom
Conversation
When a client (e.g. a phone-app user) disconnects and is inactive for more than INACTIVE_CLIENT_TIMEOUT (1 hour), the eviction routine marks their room_client_sync row with last_activity=0 and removes them from the in-memory ACL. Previously, when that client reconnected and was re-added to the ACL, the sync loop would see last_activity=0 and permanently skip them — so any messages that arrived while they were offline were never delivered. Fix: treat last_activity==0 as a reconnect signal rather than a permanent skip. The sync loop now restores the client's sync state (preserving sync_since so already-seen messages are not replayed, resetting push_failures and pending_ack_crc) and falls through to the normal unsynced-message check so the backlog is delivered immediately. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds a new `pubkey_aliases` table so admins can assign human-readable display names to client public keys that never appear in the adverts table (e.g. MeshCore phone-app users whose device does not broadcast LoRa adverts). - sqlite_handler.py: `pubkey_aliases` table created on DB init plus migration 12 for existing databases; CRUD helpers set/get/delete/list - storage_collector.py: `get_node_name_by_pubkey()` now checks aliases first (highest priority), then falls back to the adverts table as before; new wrapper methods set/delete/list aliases forwarded to the DB handler - api_endpoints.py: new GET /api/pubkey_aliases and POST|DELETE /api/pubkey_alias endpoints; GET /api/resolve_pubkey_names batch lookup for the UI poll cycle; `acl_clients` response now includes `node_name` resolved via the same lookup chain Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…nd API endpoint 53 tests across four files: - test_sqlite_pubkey_aliases: CRUD against real in-memory SQLite - test_storage_collector_name_resolution: alias-first priority chain - test_room_server_reconnect: eviction/reconnect sync loop behaviour - test_resolve_pubkey_names_endpoint: batch name resolution endpoint Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Clients retry a message when they don't receive an ack fast enough (common for distant nodes with weak signal). Without deduplication every retry lands as a separate message in the room. Add an in-memory `_recent_posts` cache keyed on (pubkey_hex, message_text). Any identical (author, text) pair arriving within DEDUP_WINDOW_SECS (30s) is silently dropped. Cache is evicted when it grows past 500 entries to bound memory usage. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
5 tests covering TestAddPostDedup: - First post is stored normally - Identical (author, text) within 30s window is dropped - Same text accepted again after the window expires - Different text from the same author always accepted - Same text from different authors both stored independently Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The MeshCore client retransmits unacknowledged messages up to 3 times.
Each retransmission carries the SAME sender_timestamp (the original send
time encoded in the packet by the firmware), making it a reliable identity
for deduplication regardless of how the text bytes decode.
Previously the dedup key was (author, message_text) + a 30 s time window.
This failed in two ways:
1. The client's retry interval can exceed 30 s, so the third attempt
often slipped through.
2. pymc_core decodes invalid trailing bytes as U+FFFD (□) via
errors="replace". If one copy decodes as "Test" and a retry as
"Test□" the text keys differ and the duplicate is stored anyway.
Changes:
- pymc_core text handler now exposes sender_timestamp, attempt, and
txt_type in packet.decrypted so callers can use them.
- repeater text.py extracts the client timestamp and prefers it over
int(time.time()) when calling add_post(); also strips U+FFFD from
message text alongside the existing null-byte strip.
- room_server add_post() uses (author_hex, sender_timestamp) as the
primary dedup key. Falls back to (author_hex, normalised_text) +
DEDUP_WINDOW_SECS for the web-API path where sender_timestamp is 0.
- Tests updated: 7 → 7 cases covering both the timestamp path and the
text-window fallback, including the mixed-text retry scenario.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…hanges The previous commit introduced a timestamp-based dedup path that required pymc_core to expose sender_timestamp in packet.decrypted. That dependency is unnecessary — the simpler fix is to normalise the message text before comparing, which collapses all retransmission variants to the same string. pymc_core decodes invalid trailing bytes as U+FFFD (□) via errors="replace". The firmware null-terminates C strings. Either way, all retransmissions of the same message should produce identical normalised text. Changes: - text.py: extend rstrip to also strip U+FFFD (□) alongside existing \x00 strip. Comment explains why: retransmissions can have trailing replacement chars. - room_server.py: revert to simple (author, normalised_text) + time-window dedup. No sender_timestamp involvement; works with any pymc_core version. - tests: restore 5+2 test layout covering dedup, window expiry, per-author isolation, and web-API (sender_timestamp=0) path. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Four improvements to the room server, each independently useful:
1. Fix: queued messages not delivered after client eviction (
room_server.py)When a client's
last_activityis0(evicted by the inactivity timer), reconnecting clients were silently skipped by the sync loop — theirpush_failurescounter had already hitMAX_PUSH_FAILURESand the loop checked that guard before checking for a reconnect. Messages queued during their absence were never delivered.Fix: detect reconnect first (active client +
last_activity == 0), resetpush_failuresand restorelast_activity, then continue to the normal push path.2. Fix: deduplicate client retransmissions (
room_server.py)Clients retry a message when they don't receive an ack fast enough — common for distant nodes with weak signal. Without deduplication every retry lands as a separate message in the room.
Add an in-memory
_recent_postscache keyed on(pubkey_hex, message_text). Any identical (author, text) pair arriving withinDEDUP_WINDOW_SECS(30 s) is silently dropped. Cache is evicted when it grows past 500 entries to bound memory usage. Server-authored messages (web UI posts) bypass the check via the existingallow_server_authorflag.3. Feat: pubkey alias system (
sqlite_handler.py,storage_collector.py,api_endpoints.py)Phone-app users connecting via the room server protocol never broadcast LoRa adverts, so their messages always appear as an unresolved pubkey. This adds a
pubkey_aliasestable so admins can manually assign display names.pubkey_aliasestable created on DB init; migration 12 for existing databasesget_node_name_by_pubkey()onStorageCollectorchecks aliases first (highest priority), then falls back to the adverts tableGET /api/pubkey_aliases— list all aliasesPOST /api/pubkey_alias— set alias{pubkey, alias}DELETE /api/pubkey_alias?pubkey=...— remove aliasGET /api/resolve_pubkey_names?pubkeys=a,b,c— lightweight batch lookup for the UI poll cycle, returns{pubkey: name|null}mapGET /api/acl_clientsresponse now includesnode_name(alias → advert fallback) for each connected client4. Test coverage (
tests/)58 new tests across 4 files:
test_sqlite_pubkey_aliases.pytest_storage_collector_name_resolution.pytest_room_server_reconnect.pytest_resolve_pubkey_names_endpoint.pyTests use
importlib.util.spec_from_file_locationto load modules directly, bypassing__init__.pychains that pull in unavailable hardware dependencies.UI changes
Companion UI PR: tjdownes/pyMC-RepeaterUI#1 (draft, targeting pyMC-dev/pyMC-RepeaterUI)
Per discussion with @rightup, built UI assets are not included here — the UI source PR should be reviewed and built separately.
🤖 Generated with Claude Code