fix(dev): make 'make dev-up' work — readable temp compose files + HAKeeper bootstrap

[fengttt]

What

Two fixes that let the etc/docker-multi-cn-local-disk stack (make dev-up) actually start a working multi-CN cluster. The first failure masked the second, so both are needed.

1. Temp compose fragments written to /tmp (start.sh)

start.sh generated the default-limits and extra-mounts compose fragments under /tmp. A snap-confined Docker daemon cannot read /tmp, so docker compose failed before any container started:

open /tmp/docker-compose.default-limits.<pid>.yml: no such file or directory

Now the fragments are written in the project directory (under $HOME, which the daemon can access) using relative paths, so docker compose keeps it as the project directory. Also fixes a latent cleanup bug: the two per-block trap ... EXIT handlers clobbered each other (the second overwrote the first), leaking the default-limits file whenever extra mounts were used — both temp files are now tracked in one list and removed by a single trap.

2. HAKeeper never bootstrapped (generate-config.sh)

The generated log.toml had no [logservice] section. Each node is started separately with -cfg log.toml (not -launch), so nothing ever bootstrapped HAKeeper and the log store panicked on first start:

panic: shard not bootstrapped

The launch stacks avoid this only because -launch builds the bootstrap config programmatically; no checked-in toml sets bootstrap-cluster. (This was invisible until fix #1, since compose always died at the /tmp step before a container ran.)

generate-config.sh now emits a [logservice] section that gives mo-log a UUID and bootstraps a single-replica HAKeeper (1 log shard, 1 tn shard, 1 replica, init member 131072:<uuid>), matching pkg/embed/cluster.go and logservice.DefaultConfig. Addresses advertise the mo-log hostname while listening on 0.0.0.0, and gossip-allow-self-as-seed is set because Fill() (unlike DefaultConfig) does not default it — required for a lone seed node. The *.toml files are gitignored (generated), so the fix lives in the generator.

Verification

From a clean state (make dev-clean && make dev-up), validated through the generated config path (tomls removed → regenerated by generate-config.sh → fresh bring-up):

• all 5 containers up, mo-log reports healthy, zero shard not bootstrapped panics
• create / insert / select sum() round-trips through the proxy on port 6001

🤖 Generated with Claude Code

[qodo-code-review]

Qodo reviews are paused for this user.

Troubleshooting steps vary by plan Learn more →

On a Teams plan?
Reviews resume once this user has a paid seat and their Git account is linked in Qodo.
Link Git account →

Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center?
These require an Enterprise plan - Contact us
Contact us →

[Copilot]

Pull request overview

This PR fixes the local multi-CN Docker Compose dev stack (etc/docker-multi-cn-local-disk, invoked via make dev-up) so it can reliably start a functional cluster by (1) generating compose override fragments in a Docker-readable location and cleaning them up correctly, and (2) ensuring the logservice boots HAKeeper when started via -cfg log.toml (not -launch).

Changes:

• Write temporary docker-compose fragments next to docker-compose.yml (instead of /tmp) and clean them up via a single trap to avoid leaking temp files.
• Extend generate-config.sh to emit a [logservice] section (including BootstrapConfig) in log.toml so a single-node HAKeeper is bootstrapped for this dev stack.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
etc/docker-multi-cn-local-disk/start.sh	Writes compose fragment temp files in the project directory and centralizes cleanup so docker compose works with snap-confined Docker.
etc/docker-multi-cn-local-disk/generate-config.sh	Adds logservice + bootstrap configuration to generated log.toml so HAKeeper is initialized and log store doesn’t panic on first start.