2026 Remote Mac Pull Stability FAQ: Git / Homebrew / npm Retry & Timeout

Three main reasons:

• Long RTT and packet loss. Cross-border or remote links add latency and loss. Default client timeouts are short; a single slow segment can abort the transfer.
• Small buffers and no retry. Git’s default http.postBuffer is small for large pushes; HTTP clients often don’t retry on transient errors. Homebrew and npm use similar defaults.
• No connection reuse. Repeated SSH connections (e.g. for Git over SSH) renegotiate each time. Without ControlPersist, each clone or fetch pays full connection cost and is more exposed to mid-transfer drops.

CI runners on a remote Mac amplify this: many jobs clone and install in parallel, so timeouts and retries become critical for stability. Cross-border teams often see the worst of it when the Mac node and the Git/npm/Homebrew upstream are in different regions.

Use these globally or per-repo to reduce timeouts and improve retry behavior.

git config --global http.postBuffer 524288000
git config --global http.lowSpeedLimit 1000
git config --global http.lowSpeedTime 60

Optional HTTP timeout (seconds): git config --global http.version HTTP/1.1 can help on some proxies. For a hard timeout: git config --global http.lowSpeedTime 120 (increase if links are very slow). In CI, set these via git config in the job or in a pre-installed global config on the runner image.

SSH connection reuse (ControlPersist): Add to ~/.ssh/config for your host (e.g. Host github.com):

Host github.com
  ControlMaster auto
  ControlPath ~/.ssh/sockets/%r@%h-%p
  ControlPersist 600

Create the directory: mkdir -p ~/.ssh/sockets. New connections reuse the master for 600 seconds, reducing reconnects and mid-clone failures.

Homebrew. Set retries and (optionally) a faster core remote:

export HOMEBREW_CURL_RETRIES=5
export HOMEBREW_CORE_GIT_REMOTE="https://github.com/Homebrew/homebrew-core.git"
# Optional: mirror
# export HOMEBREW_BOTTLE_DOMAIN="https://mirror.example.com"

Put exports in ~/.zshrc or your CI env. Increase HOMEBREW_CURL_RETRIES on flaky networks.

npm. Retries and timeouts (e.g. in .npmrc or env):

fetch-retries=5
fetch-retry-mintimeout=20000
fetch-retry-maxtimeout=120000
fetch-timeout=300000

Times are in milliseconds. For CI, a registry mirror or proxy often matters more than tuning these alone; combine with the table below. Put these in project .npmrc or in the CI environment so every run uses the same retry and timeout policy.

Choose by use case, ops capacity, and cost.

For remote Mac CI: start with a public mirror or proxy if your runner is cross-border; move to a self-hosted mirror if you need maximum stability and can run sync jobs. See our Git mirror vs proxy guide and cross-border pull acceleration for step-by-step config.

• Increase Git http.postBuffer, set lowSpeedLimit / lowSpeedTime, and enable SSH ControlPersist for connection reuse.
• Set Homebrew HOMEBREW_CURL_RETRIES and optionally a core remote or bottle mirror; set npm fetch-retries and timeouts (and a registry mirror if needed).
• Pick mirror vs proxy vs self-hosted from the table above; combine with shallow clone or cache where possible (see our cache strategy post).
• Quick checklist: Git buffer and lowSpeed* set; SSH ControlPersist in place; Homebrew retries and npm retries/timeouts configured; mirror or proxy chosen and applied. Re-run a failing clone or install after each change to confirm.