Why

Pid2's databricks.ts invokes databricks sync to upload Repl files to a Databricks Apps workspace. We've been hitting 502s mid-sync that surface as fatal errors because the upstream SDK's retry predicate covers 429/504 but not 502/503. We also want the option to dial in-flight upload concurrency.

The fork at replit/databricks-cli adds two flags to databricks sync (and databricks bundle sync):

• --retry-timeout (default 30s) — per-call deadline for retrying transient gateway errors (HTTP 502/503/504), implemented with the SDK's retries.Poll.
• --concurrency (default 5) — replaces the hardcoded 20 in-flight request limit.

Fork branch: https://github.com/replit/databricks-cli/tree/replit-main
Upstream PR (merged): replit/databricks-cli#1
Slack thread: https://replit.slack.com/archives/C0A2Z9042FR/p1778082184841149

What changed

• Repoint src from databricks/cli v0.286.0 → replit/databricks-cli replit-main at commit 4928653c. The branch is upstream v0.290.2 + the 1 Replit commit. Version string is 0.290.2-replit-4928653c.
• Update source hash and vendorHash. Vendor hash is identical to nixpkgs' v0.290.2 entry (the Replit commit doesn't add new Go dependencies).
• Update meta.homepage to the fork; keep meta.changelog pointing at upstream tags.
• Pin to v0.290.2 (the same release nixpkgs is on at https://github.com/NixOS/nixpkgs/blob/master/pkgs/by-name/da/databricks-cli/package.nix) so the checkFlags skip list stays identical to nixpkgs', mirroring established practice. v0.291+ adds an SDK host-metadata resolver that's awkward to test in the nix sandbox; bumping past it is a separate change.

Test plan

$ nix build '.#activeModules.databricks-cli' '.#activeDeploymentModules.databricks-cli'
# both targets build clean, including the upstream test phase

$ databricks --version
Databricks CLI v0.290.2-replit-4928653c

$ databricks sync --help | grep -E "concurrency|retry-timeout"
      --concurrency int          maximum number of concurrent in-flight requests during sync (default 5)
      --retry-timeout duration   per-call deadline for retrying transient gateway errors (HTTP 502/503/504) (default 30s)

135 test packages run in the nix sandbox during nix build, including libs/sync (covers the new retry helper) and cmd/sync (covers the new flags). 0 failures. The checkFlags skip list is unchanged from the original PR (#477) and mirrors nixpkgs upstream.

Rollout

• This is fully backward and forward compatible

The new flags have sane defaults, so existing callers (pid2's databricks sync ... invocation) get the new retry behaviour and concurrency=5 automatically without any pid2 change. To revert, change the pin back to upstream databricks/cli v0.286.0 with the original hashes.

Revertibility

Safe to revert. Pure module pin change with no migrations or persistent state.

~ written by Zerg 👾 (wp-7918444a)