Server v1.15.10 crashes repeatedly due to massive memory leak and file watcher binding failure #29177

Environment

• Version: 1.15.10
• Binary: Bun-compiled ELF binary at ~/.opencode/bin/opencode
• OS: Linux, x86-64, 14 GB RAM
• Start command: opencode serve --hostname <ip> --port 4096

Symptoms

The server crashes repeatedly (11+ restarts since May 22, 32 ERROR lines in one session). Two primary issues:

1. Massive memory leak

The process virtual memory grows unbounded:

VmPeak: 135413388 kB
VmSize: 74711184 kB  (and growing ~65 MB/minute)
VmData: 65349732 kB
VmRSS:     393488 kB

RSS is only ~400 MB, but virtual address space balloons to 71+ GB and keeps growing until the process is killed. This happens irrespective of workload.

2. File watcher native binding fails to load

Every session starts with this error — the binary references a Bun runtime artifact path that does not exist:

ERROR service=file.watcher error=/$bunfs/root/watcher-0pmc4yjg.node: cannot open shared object file: No such file or directory failed to load watcher binding

3. Share service unreachable

ERROR service=share-next ... share flush failed
cause: ConnectionRefused to https://opncd.ai/api/share/...

This repeats many times per session (likely a retry storm contributing to memory pressure).

4. Session processing aborted on shutdown

ERROR service=session.processor session.id=ses_1a77f9872ffeBYegxBfc4c9pdW messageID=msg_e5c018199001GscPkBEJy1GS3d error=Aborted process
ERROR service=session.processor session.id=ses_1a9651492ffe83maUzVPdKSjia messageID=msg_e5899363b001use0oxgd3tVbPU error=Aborted process

Logs

~/.local/share/opencode/log/

Expected behavior

• Server should not allocate 65 GB of virtual memory for normal operation
• File watcher should gracefully degrade if the native binding cannot load
• Share sync failures should not produce cascading errors

Please investigate the memory leak (likely in the Bun runtime or file watcher binding) and the missing native module path at /$bunfs/root/watcher-0pmc4yjg.node.