Auto-updater memory leak crashed 12-hour session (v2.1.76) — 13.81 GB committed, Bun panic

[kolkov]

Summary

Claude Code v2.1.76 crashed overnight with a Bun panic ("Illegal instruction") after accumulating 13.81 GB of committed memory during a 12-hour session. The crash appears related to the auto-updater memory leak that was fixed in v2.1.77:

"Fixed auto-updater starting overlapping binary downloads when the slash-command overlay repeatedly opened and closed, accumulating tens of gigabytes of memory"

Crash Details

Bun v1.3.11 (a2743b6d) Windows x64 (baseline)

Elapsed: 44576556ms (~12.4 hours)
RSS: 2.58GB
Peak: 2.58GB
Commit: 13.81GB          ← auto-updater leak
Faults: 27235766         ← 27M page faults
Machine: 42.62GB RAM

panic(thread 76628): Illegal instruction at address 0x7FF641918940

Crash report: bun.report/1.3.11/e_1a2743b6mgkgE+hogCg0ku4C6gst6Bw+ps4Cqutt6Birwt6B+57s6B4kz+6Bwnpm0C0rln0Ds0w2rDy25urD2glrrDsvtmrDqkomrD667k8Cum0k8Cyxi61Cgi0i0Cgwk70Cyk8z

What Happened

1. Started a Claude Code session on v2.1.76 in the evening
2. Left it running overnight (agentic task in progress)
3. After ~12 hours, Bun crashed with "Illegal instruction" — session lost
4. Memory had grown to 13.81 GB committed (on a machine with 42 GB RAM)

Root Cause

The v2.1.77 changelog confirms this was a known bug:

• Auto-updater started overlapping binary downloads when the slash-command overlay opened/closed
• Downloads accumulated in memory, growing to tens of gigabytes
• Eventually Bun ran out of addressable memory and hit an illegal instruction

Trigger

We had just removed CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC from settings to enable the 1M context window (which requires this traffic to be enabled — see #34143). This also re-enabled the auto-updater, which triggered the memory leak.

Users are forced to choose between:

• 1M context (requires traffic enabled) → auto-updater leak crashes sessions
• Stability (disable traffic) → stuck at 200K context

Environment

• Claude Code: v2.1.76 (now auto-updated to v2.1.77)
• OS: Windows 10 Home 10.0.19045
• Model: Opus 4.6 with 1M context
• Plan: Claude Max
• RAM: 40 GB

Impact

• 12 hours of agentic work lost — session cannot be recovered after Bun crash
• No warning before crash — memory grew silently in the background
• The auto-updater should not be downloading anything during an active session

Suggestion

Even with the v2.1.77 fix for overlapping downloads, the auto-updater should:

1. Never download during an active session — wait for idle or session end
2. Have a memory budget — abort if download memory exceeds a threshold
3. Not be coupled with 1M context — CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC should only disable telemetry, not block context window features

Related: #33949 (root cause analysis of streaming hangs), #34143 (1M context not working with traffic disabled)

[github-actions]

Found 2 possible duplicate issues:

1. Memory leak: self-upgrade fetch Response body not released, causing 45GB+ external memory accumulation (Bun ArrayBuffer finalizer) #33874
2. [BUG] CRAZYE MEMORY LEAK. Can't use Claude Code. --update doesn't work as well #34424

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[kolkov]

Not a duplicate:

• Memory leak: self-upgrade fetch Response body not released, causing 45GB+ external memory accumulation (Bun ArrayBuffer finalizer) #33874 (closed) — describes the memory leak mechanism (fetch Response body not released). Our issue documents the real-world impact: 12-hour overnight session lost, and the fact that this bug is coupled with 1M context activation — users must remove CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC to get 1M context (Opus 4.6 shows 200K context instead of 1M on Claude Max plan (v2.1.75) #34143), which also enables the leaky auto-updater.
• [BUG] CRAZYE MEMORY LEAK. Can't use Claude Code. --update doesn't work as well #34424 — reports a general memory leak. Our issue identifies the specific trigger (auto-updater + traffic setting) and proposes fixes: don't download during active sessions, don't couple telemetry with context window features.

The unique contribution here is the 1M context ↔ auto-updater coupling problem: you can't have 1M context without also enabling the auto-updater that crashes your sessions.

[TheStack-ai]

If anyone is experiencing leftover processes consuming memory after Claude Code sessions end, I built a small CLI tool that automatically cleans up orphaned child processes (MCP servers, sub-agents, headless browsers, etc.) that don't exit when the parent session dies.

It only targets processes whose parent is already dead — active sessions are never touched.

npx z-clean init

https://github.com/whynowlab/zclean

This won't fix the auto-updater memory leak itself (that's a Claude Code internal issue), but it helps with the broader pattern of orphaned processes accumulating across sessions.

[kolkov]

Update: Second Bun crash — this time WITHOUT auto-updater

Another Bun mimalloc panic on v2.1.79, with DISABLE_AUTOUPDATER=1 set. Auto-updater is NOT the only cause.

Elapsed: 85,278,850ms (23.7 hours)
RSS: 1.73GB | Peak: 1.93GB | Commit: 15.40GB | Faults: 67,347,099

panic(thread 59820): Illegal instruction at address 0x7FF7A3A98940

Comparison:

Metric	Crash #1 (v2.1.76)	Crash #2 (v2.1.79)
Session duration	12.4 hours	23.7 hours
Committed memory	13.81 GB	15.40 GB
Page faults	27M	67M
Auto-updater	Enabled (leak)	Disabled
Root cause	auto-updater + mimalloc	mimalloc only

The auto-updater fix in v2.1.77 helped, but Bun's mimalloc allocator itself leaks memory during long sessions — roughly 0.6-1 GB/hour of committed memory growth. After 12-24 hours, mimalloc's scavenger hits an inconsistent state → "Illegal instruction" → crash.

Before the crash: the session showed "Gesticulating... 13m 27s" — 13 minutes of hang before the panic. The streaming watchdog didn't help because this is a runtime crash, not an SSE stall.

Bottom line: Bun runtime is not suitable for long-running interactive sessions. Sessions over 12 hours are at risk of mimalloc crash regardless of other fixes.

[kolkov]

Update: Switched from native (Bun) to npm (Node.js) — dramatic stability improvement

After 6 Bun crashes in 4 days, 26 GB commit from 7 processes, IDE crashes, and system deadlocks — we decided to switch back from the native installer (Bun 1.3.11) to the npm package (Node.js v22.17.0). Same cli.js, same settings, different runtime.

# How we switched (keep your settings!)
cp -r ~/.claude ~/.claude-backup          # backup first
rm ~/.local/bin/claude.exe                # remove native binary (NOT claude uninstall!)
npm install -g @anthropic-ai/claude-code  # install npm version

Results after first hours on Node.js

Memory — night and day:

Metric	Bun (native)	Node.js (npm)
Commit per process	1-15 GB (grows forever)	300-500 MB (stable)
Commit ≈ RSS ratio	Commit 10-15x larger than RSS	Commit ≈ RSS
5 processes total	26 GB commit	1.6 GB commit
System RAM used	70%+ (constantly swapping)	47% (21 GB free)
Memory over time	Grows 0.6-5 GB/hour	Flat or decreasing

Stability:

• 0 mimalloc crashes (impossible on Node.js)
• 0 "Tool bash not found" (no memory corruption)
• 0 system deadlocks
• 0 IDE crashes from memory pressure
• V8 GC actually returns memory to OS (mimalloc never did)

What still remains (cli.js / server-side):

• Streaming stalls — still happen (server-side issue)
• Tool result loss — insertMessageChain() for-loop still non-atomic

What ALSO disappeared (we thought it was Ink, it was Bun):

• Scroll resets (Full reset (scrollback changes)) — 316 per session on Bun → 0 on Node.js
• Input field rendering corruption — text appearing below prompt, garbled input — gone
• Scrolling history back and forth works smoothly — no jumps, no discomfort
• We originally attributed all of this to Ink's renderer (Terminal Scrolling Uncontrollably During Claude Code Interaction #3648, 👍694). Debug logs prove it was Bun/mimalloc corrupting render state.

The theory: Bun/mimalloc was THE root cause

We originally identified 3 separate root causes (#33949). But switching runtimes suggests many symptoms may have been one root cause — mimalloc memory corruption:

• "SSE connection dies" → corrupted HTTP client state
• "Tool result lost" → corrupted JSONL write buffer (partially — for-loop bug is real too)
• "Scroll jumps" → corrupted render state
• "Tool bash not found" → corrupted tools registry
• "21-thread deadlock" → corrupted event loop / mutex state
• "System freeze" → mimalloc scavenger blocking all threads

All of these disappear or dramatically reduce on Node.js/V8.

Recommendation

If you're experiencing crashes, memory bloat, or instability — try the npm install. You lose ~200ms startup time and auto-updates, but gain a runtime that's been stable for 15+ years. The native Bun installer is the default, but the npm package is still fully supported (v2.1.83).

Note: You'll see an "npm deprecated" banner on each launch — it's just a message, ignore it.