drove-cli

name

drove-cli

description

Operate, inspect, and manage Drove container-orchestrator deployables (applications, local services, tasks, app instances, executors, and cluster) using the `drove` CLI tool. Use this skill whenever the user asks to deploy, scale, restart, suspend, destroy, list, inspect, or tail logs for anything on a Drove cluster — or to manage Drove CLI configuration (endpoints, credentials). Activate when you see words like: drove, app deploy, scale app, local service, localservice, lsinstance, drove task, container orchestrator, drove cluster, drove endpoint, drove executor, drove config. Even if the user just pastes an app-id or says "check on my app", use this skill.

drove-cli Skill

You are an expert operator of Drove, an open-source container orchestrator, using the drove CLI.

---

Step 0 — Check and Install drove-cli

Before running any drove command, verify the CLI is available:

which drove 2>/dev/null || python3 -c "import drovecli" 2>/dev/null

If missing, install it:

pip install drove-cli

After installation verify:

drove --version

If drove is still not on PATH after pip install, try python3 -m drove as a fallback invocation prefix.

---

Step 1 — Load the Full Command Reference

Run this once to bring every command and sub-command into context:

drove --full-help

This prints the complete, always-up-to-date help for every command and sub-command in one shot. Use it to look up flags, argument names, and available options — no need to memorise them.

Full documentation: https://phonepe.github.io/drove-orchestrator/cli/ Refer to this whenever the built-in help isn't enough or something is unclear.

---

Step 2 — Configure Cluster Connection

drove reads connection details from ~/.drove or from env vars / CLI flags.

Option A – CLI flags (one-off)

drove -e http://my-cluster:10000 -u admin -p secret cluster ping

Option B – Environment variables (CI / scripts)

export DROVE_ENDPOINT=http://my-cluster:10000
export DROVE_USERNAME=admin
export DROVE_PASSWORD=secret

Option C – Config file ~/.drove

drove config init --endpoint http://my-cluster:10000 --name prod \
                  --username admin --password secret
drove config use-cluster prod      # switch active cluster
drove config get-clusters          # list saved clusters

Always verify connectivity before running other commands:

drove cluster ping     # → "Ping successful"

---

Step 3 — Resource Types

Resource	Plugin	Description
Application	apps	Long-running containerised services
App Instance	appinstances	Individual running containers of an app
Local Service	localservices	Services pinned to specific executor nodes
LS Instance	lsinstances	Individual containers of a local service
Task	tasks	Short-lived one-shot jobs
Executor	executor	Worker nodes in the cluster
Cluster	cluster	Cluster-level health and meta-commands
Describe	describe	Deep-dive JSON view of any resource

---

Step 4 — Common Workflows

Deploy a new app

drove apps create my_app.json
drove apps scale  MY_APP-1 3 --wait
drove apps summary MY_APP-1

Rolling restart

drove apps restart MY_APP-1 --parallelism 1 --timeout 10m --wait

Tear down an app

# App must be suspended before it can be destroyed
drove apps suspend MY_APP-1 --wait
drove apps destroy MY_APP-1

Tail logs from a running instance

drove appinstances list MY_APP-1          # find instance IDs
drove appinstances tail MY_APP-1 AI-abc123 --log output.log

Run a one-shot task and monitor it

drove tasks create my_task.json
drove tasks show  MY_APP T0042
drove tasks tail  MY_APP T0042 --log output.log

Activate a local service

drove localservices create   my_service.json
drove localservices conftest MY_SVC-1      # validate with one instance first
drove localservices activate MY_SVC-1
drove localservices summary  MY_SVC-1
drove lsinstances list       MY_SVC-1       # find local service instance IDs
drove lsinstances tail       MY_SVC-1 SI-abc123 --log output.log

---

Gotchas

• drove always exits 0, even on API errors — check stdout for "not found" or "error" rather than relying on the return code in scripts
• tasks list shows only running tasks — use tasks show <source> <id> for completed or stopped tasks
• Local service healthy state is ACTIVE (not RUNNING)
• An app must be in MONITORING state (i.e. suspended) before it can be destroyed — suspend it first
• For SSL-terminated clusters with a self-signed cert, pass --insecure / -i