[planning] E2EE / call-key cluster KE-1→4: OTK upload conflict storm, EC MissingKey, malformed algorithm, MatrixRTC delayed-event timeouts — capture first, do not spot-fix #201

New Issue

2026-09-17T23:25:31-04:00

jared commented

2026-09-17 23:25:31 -04:00

Migrated from LOTUS_TODO.md on 2026-09-17 (the file is now reference-only).

🧨 Encryption / E2EE — ⚠️ EXTREME COMPLEXITY · 🧠 PLANNING SESSION REQUIRED

Observed live in prod 2026-06-30 during a 2-person Element Call (E2EE). These span client rust-crypto (matrix-js-sdk@41.7.0) ↔ Synapse ↔ EC MatrixRTC E2EE and are interrelated — do NOT spot-fix. Capture first: run Settings → Developer Tools → Crypto Diagnostics during the next affected call + a synapse-side trace before any fix. (Full runbook was in LOTUS_E2EE_INVESTIGATION.md, now in git history.) None are caused by the EC fork work.

KE-1 — OTK upload conflict storm (CRITICAL, root-cause candidate). POST /keys/upload returns 400 M_UNKNOWN: One time key … already exists continuously — the rust-crypto store and Synapse have diverged OTK state (upstream matrix-rust-sdk#5200, OPEN: on the 400 the SDK never marks the request sent → re-uploads forever; not fixed in 41.7.0). Leading web trigger: cinny never calls navigator.storage.persist(), so the IndexedDB crypto store is evictable while the localStorage session survives → device resurrects with a blank store. Buildable preventive fix (no call needed): request persistent storage on login (+ optional multi-tab guard + a 400-loop→recovery prompt). Healing an already-diverged device still needs a clean logout+login.
KE-2 — EC media keys not arriving/decrypting → audio/video cut out (CRITICAL). MissingKey … for participant, unexpected encrypted to-device io.element.call.encryption_keys. Almost certainly downstream of KE-1 (broken Olm sessions). This is the "friend's audio cuts out" symptom.
KE-3 — Timeline decrypt error: missing algorithm field (HIGH). rust-crypto can't parse a malformed/legacy encrypted event — capture the offending event id + raw content.
KE-4 — MatrixRTC delayed-event / membership timeouts (MEDIUM-HIGH). Restart delayed event timed out, repeated msc4157.update_delayed_event — may be partly HS responsiveness; correlate with synapse latency. Same planning session (shares the call-reliability surface).

Manual capture steps: [QA] #184 (§O8).

_Migrated from `LOTUS_TODO.md` on 2026-09-17 (the file is now reference-only)._ ### 🧨 Encryption / E2EE — ⚠️ EXTREME COMPLEXITY · 🧠 PLANNING SESSION REQUIRED Observed live in prod 2026-06-30 during a 2-person **Element Call** (E2EE). These span client rust-crypto (`matrix-js-sdk@41.7.0`) ↔ Synapse ↔ EC MatrixRTC E2EE and are **interrelated** — do NOT spot-fix. **Capture first:** run **Settings → Developer Tools → Crypto Diagnostics** during the next affected call + a synapse-side trace before any fix. (Full runbook was in `LOTUS_E2EE_INVESTIGATION.md`, now in git history.) None are caused by the EC fork work. - **KE-1 — OTK upload conflict storm (CRITICAL, root-cause candidate).** `POST /keys/upload` returns `400 M_UNKNOWN: One time key … already exists` continuously — the rust-crypto store and Synapse have **diverged OTK state** (upstream `matrix-rust-sdk#5200`, OPEN: on the 400 the SDK never marks the request sent → re-uploads forever; **not** fixed in 41.7.0). Leading web trigger: cinny never calls **`navigator.storage.persist()`**, so the IndexedDB crypto store is evictable while the `localStorage` session survives → device resurrects with a blank store. **Buildable preventive fix (no call needed):** request persistent storage on login (+ optional multi-tab guard + a 400-loop→recovery prompt). Healing an already-diverged device still needs a clean logout+login. - **KE-2 — EC media keys not arriving/decrypting → audio/video cut out (CRITICAL).** `MissingKey … for participant`, unexpected encrypted to-device `io.element.call.encryption_keys`. Almost certainly downstream of KE-1 (broken Olm sessions). This is the "friend's audio cuts out" symptom. - **KE-3 — Timeline decrypt error: missing `algorithm` field (HIGH).** rust-crypto can't parse a malformed/legacy encrypted event — capture the offending event id + raw content. - **KE-4 — MatrixRTC delayed-event / membership timeouts (MEDIUM-HIGH).** `Restart delayed event timed out`, repeated `msc4157.update_delayed_event` — may be partly HS responsiveness; correlate with synapse latency. Same planning session (shares the call-reliability surface). Manual capture steps: [QA] #184 (§O8).