TELCORE-248 / TEL-6872: Cherry-pick elastic jitter buffer fixes from SignalWire PR #2337 - #539
Merged
tajamulTelnyx merged 7 commits intoJul 16, 2026
Conversation
dev-ryanc
requested review from
baloeng,
damirn,
minhtuan1407-telnyx and
tajamulTelnyx
March 13, 2026 12:31
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
minhtuan1407-telnyx
approved these changes
Mar 13, 2026
damirn
approved these changes
Mar 13, 2026
damirn
left a comment
damirn
left a comment
There was a problem hiding this comment.
can't complain on fixes from upstream ;)
|
@dev-ryanc make sure you grab the second commit 60f8f34 This is preventing a race condition that can take place is someone extract the statistics in a hangup hook |
minhtuan1407-telnyx
approved these changes
May 5, 2026
minhtuan1407-telnyx
left a comment
minhtuan1407-telnyx
left a comment
There was a problem hiding this comment.
🔬 Analyst review — TEL-6872 elastic JB cherry-pick
Verdict: APPROVE (with non-blocking notes)
Changes verified
MAX_DROPOUT=3000guard incheck_jb_size()— hides "old" nodes only whenudelta ∈ (1, 3000). Prevents seq-wraparound hiding valid packets. ✅decrement_seq()ordering fix —last_target_seqnow set AFTER decrementingtarget_seq. Old code left both fields equal after expand, breaking PLC seq emission. ✅expand_frame_lentracking +RESET TOO EXPANDED— increments on PLC expand, decrements on acceleration drop, resets JB ifexpand_frame_len > max_frame_len. Prevents unbounded growth. ✅MAX_CONSECUTIVE_MISS=100safety valve — falls back to non-elastic + reset after 100 consecutive misses. ✅!jb->elasticgate on BUFFERING path — elastic JB serves packets without waiting to refillframe_len. ✅consecutive_missreset on successful read +buffering_skipcounter. ✅
Non-blocking concerns
- Subset of unmerged upstream. SignalWire PR signalwire#2337 is still OPEN. It also adds:
switch_jb_export_stats(),switch_rtp_get_jitter_buffer_for_stats(),switch_core_media_export_jb_stats()(flush JB stats before hangup hooks), and a multi-tier forced/fast acceleration path (packets_in_buffer >= max_frame_len). Theexpand_frame_lencap is reactive (after PLC); upstream's proactive acceleration during reads is more robust. Recommend a follow-up ticket to track upstream. - Style:
MAX_DROPOUT/MAX_CONSECUTIVE_MISSas file-scope#definevs upstream's function-localconst int. No functional difference.
CI: Jenkins green. GH Actions cancelled (24h timeout) — not regressions. 3 existing approvals (2026-03-13).
Net: ✅ Ship it. Open follow-up for upstream re-cherry-pick or forced-acceleration port.
— 🔬 Analyst
|
@dev-ryanc i see more commits in signalwire#2337 but i also wonder why is this still not merged in upstream |
…alwire#2337 Cherry-picked fixes from signalwire#2337 (Julien Chavanton): 1. Sequence rollover bug in check_jb_size() - add MAX_DROPOUT guard to distinguish old packets from seq 65535->0 wraparound 2. decrement_seq() ordering - set last_target_seq after decrement 3. Buffering skip gate - do not block elastic mode with the complete_frames < frame_len check 4. Consecutive miss safety valve - after MAX_CONSECUTIVE_MISS (100) consecutive missed frames, disable elastic mode and reset JB
Track net buffer expansion via expand_frame_len (incremented on PLC expand, decremented on acceleration). Reset the jitter buffer when expansion exceeds max_frame_len to prevent unbounded growth during sustained packet loss. Cherry-picked from signalwire#2337 (Julien Chavanton).
…e PR signalwire#2337 Consolidate inline channel variable exports from the jitter buffer hot path into switch_jb_export_stats(), called at hangup before hangup hooks run (via switch_core_media_export_jb_stats) and as a safety net before every switch_jb_destroy() in both switch_rtp_destroy() and the KILL_JB path in read_rtp_packet(). This prevents a race condition where hangup hooks read stale/missing JB stats because the jitter buffer was already destroyed or the inline exports hadn't run recently enough. Adapted from upstream commit 0388398 — excluded fast_acceleration, forced_acceleration, and packets_in_buffer fields which don't exist in our switch_jb struct. Cherry-picked from signalwire#2337 (Julien Chavanton).
…nd boundary
[1] Reset jb->frame_len to jb->min_frame_len when disabling elastic mode
after consecutive miss safety valve triggers. Without this, the JB
stays at expanded size after elastic mode is disabled.
[2] Add SWITCH_LOG_ALERT debug log when elastic mode is disabled after
MAX_CONSECUTIVE_MISS, matching upstream.
[5] Add expand_frame_len >= max_frame_len boundary check (third else-if
branch) to catch the exact boundary case and reset via reset_error.
…signalwire#2337 1. Add switch_rtp_ready() guard in the JB poll loop to prevent polling after the session is being destroyed. 2. Add rtp_jitter_buffer_accelerate channel variable check to the tries > 20 condition, allowing JB reset to be skipped when acceleration is explicitly enabled.
Add elastic algorithm activity tracking and enhanced statistics export from upstream commit af83338: - New stats fields: grew, shrunk, grow_to_max, at_min_ms, at_max_ms, edge_since, reset_setup, reset_runtime - New jb fields: media_started, packets_total - Track frame_len edge transitions in jb_frame_inc_line() - Split reset count into setup vs runtime in switch_jb_reset() - Flush at-edge timer on set_frames() - Initialize edge_since on JB creation - Export all new metrics in switch_jb_export_stats() with a comprehensive summary log line - Set media_started and increment packets_total on each put_packet Cherry-picked from signalwire#2337 (Julien Chavanton).
Collaborator
Author
@damirn Not sure why as well, but I don't see any issues with the implementation. |
baloeng
requested changes
Jul 15, 2026
tajamulTelnyx
changed the title
Fix double-decrement in decrement_seq() — last_target_seq was over-decremented by one, shifting PLC sequence numbers an extra packet behind. Fix lock-order inversion in switch_jb_export_stats() — snapshot JB fields under mutex, unlock before channel writes to avoid holding JB mutex while acquiring channel-side locks from hangup/destroy/KILL_JB contexts. Cache rtp_jitter_buffer_accelerate channel var at function entry instead of per-iteration in the retry loop, and restore an unconditional hard retry bound (tries > 100) so the read loop cannot spin indefinitely in bad-jitter conditions.
baloeng
approved these changes
Jul 16, 2026
tajamulTelnyx
merged commit Jul 16, 2026
0ed80ca
into
telnyx/telephony/deploy-development
0 of 2 checks passed
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
6 participants
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.
Cherry-pick select bug fixes from signalwire#2337 (Julien Chavanton).
Fixes
hiding valid packets near seq 65535→0 wraparound
points to the correct value
complete_frames < frame_len buffering check
consecutive missed frames, disable elastic mode and reset JB
and reset JB when it exceeds max_frame_len
mutex, unlocks, then writes channel variables and logs (avoids lock-order
inversion with channel-side locks from hangup/destroy/KILL_JB contexts)
channel var at function entry, keep unconditional hard bound at tries > 100
Intentionally excluded
uninitialized variable issue with packet_vad, drops frames during active speech)
Testing