Skip to content

fff watcher pegs CPU on repos with nested worktrees: post-rescan directory queue runs a libgit2 status per directory #50719

Closed as not planned

Description

@rtavs101

OpenCode 1.18.32 on macOS 26.5.1 (arm64), bundled @ff-labs/fff-bun 0.9.4.

Related, different trigger: #42557 (per-event status on gitignored build output), #32511 (home-directory scanning), #43356 (crash in the same resubscribe-after-scan path).

What happens

Repo: ~7k tracked files, plus 27 untracked nested git worktrees under wt/ (~240k files). fff at the repo root indexed 245,895 files and 25,847 directories.

Two TUI sessions started at the repo root ran 5-7 h with two fff-watcher-own threads each at 100% CPU (machine load average 31, 10 GB swap). sample <pid> shows the owner thread in Repository::open → Repository::statuses / status_file → git_index_read, one call per queued directory, 10-350 ms each. A third session started inside one worktree (7,874 files) had the same threads idle.

Trigger

Reproduced with @ff-labs/fff-node 0.9.4 watching the root: creating one new worktree under wt/ produced ~8k file-create and ~1k folder-create events, 2 full rescans, and 2,375 owner-thread git_status_for_paths calls (19 s CPU) in ~40 s.

After a full rescan, rescubscribe_watcher_post_scan queues every indexed directory for the owner thread. The queue is unbounded and not deduplicated (scan.rs, background_watcher.rs in fff v0.9.4), so each rescan costs minutes of CPU per fff instance, and the queue never drains while worktrees keep being created.

A read-only experiment (fresh index, one requested rescan, then one minute of measurement):

Measurement fff 0.9.4 fff 0.11.0
CPU seconds, 10 s baseline 0.739 0.063
CPU seconds, minute after rescan 49.258 11.339
CPU seconds, final 10 s 7.347 0.035
Owner thread busy in git status waiting

fff 0.11.0 separates Subscribe from IndexNewDir, so a rescan's subscription no longer indexes every directory on macOS.

Config gaps

  • watcher.ignore reaches only the Parcel watcher, not fff.
  • fff InitOptions has no ignore/exclude option; only gitignore rules shrink the index.
  • disableWatch exists in fff-bun, but OpenCode does not pass it.
  • OPENCODE_DISABLE_FFF=1 is the only switch. With it, search falls back to ripgrep (one file query on a 7.9k-file tree: 25 ms, 0.11 s CPU).

Workaround

Gitignore the nested-worktree folder (or add it to .git/info/exclude): the root scan drops from 245,895 files to 7,140. Or set OPENCODE_DISABLE_FFF=1.

Asks

  1. Bump the bundled fff to ≥ 0.11, or backport a bounded, deduplicated post-rescan directory queue.
  2. Honour watcher.ignore when creating the fff picker.
  3. Expose fff's disableWatch behind a config key.

Activity

  1. thdxr commented on Oct 11, 2026

    @thdxr
    Member

    Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.

    — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions