GitKraken Desktop to GitHub/GitLab: Cancel behavior for long-running clone and fetch

Direct Answers

Process termination: GitKraken sends a termination signal (SIGTERM on macOS/Linux, CTRL_BREAK_EVENT on Windows) to the spawned git process when you click Cancel. The Git process then attempts to shut down its network transfer cleanly. This applies to both HTTPS and SSH remotes.

Partial-object cleanup: Git may leave a temporary .git/objects/pack/tmp_pack_* file and an incomplete FETCH_HEAD. These are harmless and are removed or overwritten on the next successful fetch/clone; GitKraken does not perform an additional cleanup step. A canceled clone leaves an empty repository directory that GitKraken deletes if the operation never recorded a valid HEAD. Partial working-tree files only appear during checkout (after fetch completes), so a canceled fetch does not affect the working tree.

Uniform client-side timeout: No single GitKraken setting applies a timeout to clone, fetch, and pull uniformly. You must configure Git global/system config (http.lowSpeedLimit, http.lowSpeedTime, http.timeout, or SSH ConnectTimeout), which affects all Git invocations on the system.

HTTPS vs SSH Transport Differences

• HTTPS: Git uses libcurl, which honors http.lowSpeedLimit (default 1000 B/s), http.lowSpeedTime (default 30 s), and http.timeout.
• SSH: The underlying SSH client (OpenSSH or PuTTY/Plink) controls timeouts via ServerAliveInterval and ConnectTimeout. GitKraken does not impose a separate deadline.

Enterprise proxy or GitHub/GitLab server-side idle timeouts (often 60–300 s) can cut the connection before Git's low-speed logic fires, which may make observed cancel or timeout behavior appear inconsistent across environments.

Version Assumptions

Behavior described reflects GitKraken 9.x/10.x releases. Versions before ~8.x occasionally left a stale index.lock on forced cancel; current releases release the lock before signaling Git.

Verification Steps

1. Start a large clone (e.g., git clone https://github.com/torvalds/linux.git), click Cancel after 5 seconds, then run:

   git -C <repo> fsck --full

   Verify no corrupt objects and that .git/objects/pack/ contains only .pack/.idx pairs, not tmp_pack_* files.
2. Repeat with an SSH remote using GIT_SSH_COMMAND='ssh -v -o ConnectTimeout=10' to confirm SSH-level timeout is respected independently of GitKraken.
3. Set low-speed thresholds to confirm Git aborts without UI cancel:

   git config --global http.lowSpeedTime 5
   git config --global http.lowSpeedLimit 1000

4. Inspect GitKraken logs (Help → Open Log Folder → gitkraken.log) for lines containing "Process terminated" or "SIGTERM sent" to confirm the signal path.

One Diagnostic Detail Needed

If you observe a hang longer than 10–15 seconds after clicking Cancel on Windows, please confirm whether you are using Git for Windows 2.40+ and whether an SSH agent (Pageant/ssh-agent) with ControlMaster is active. This determines if the delay is a known Windows signal-handling artifact or a configuration issue.