Provision a droplet per batch job and destroy it even when the job fails. Resize managed Postgres for business hours and back down at night. Snapshot the fleet by tag, catch shadow droplets minutes after they appear, and cut DNS over to the new environment as one audited operation.
Connect DigitalOcean to a workflow engine that treats the whole account as orchestrable: droplets, managed databases, block storage, load balancers, Kubernetes clusters, firewalls, and DNS zones. Kestra chains them with Ansible runs, Kubernetes deploys, Postgres jobs, and Slack signals, with retries, polling, and guaranteed teardown that doctl scripts and console clicks never give you.
DigitalOcean gives you clean primitives at a fair price. Kestra gives those primitives an operations brain: sequencing, polling, event reactions, teardown guarantees, and an audit trail across every resource type on the account.
A flow provisions a droplet with droplet.Create, injects the job through cloud-init userData, and destroys it in a finally block that runs whether the job succeeded, failed, or timed out. A crashed workload never leaves a forgotten VM billing by the hour, which is exactly what happens when the same pattern is a shell script that dies halfway.
database.Resize returns immediately while the migration runs in the background. Kestra wraps it with a database.Get health pre-check, a LoopUntil poll that only succeeds when the cluster reports online again, and per-trigger inputs so one flow scales Postgres up at 06:00 and down at 20:00. DigitalOcean has no native scheduled resize at all.
droplet.Trigger polls the account and fires one execution per droplet it has never seen, exposing its name, region, status, and creation time. That turns the public API into an event source for governance: naming policy checks, region allowlists, cost alerts, all reacting minutes after someone clicks Create Droplet outside your provisioning process.
droplet.List returns every droplet with its tags; a ForEach with bounded concurrency walks the fleet and acts per machine. Snapshot everything tagged auto-backup with a dated name, power off everything tagged ephemeral after hours. New machines join the policy by adding one tag, not by editing a server list.
domain.record.List, Delete, Create, and Get compose into a blue-green cutover: remove every stale A record for the hostname, write the low-TTL replacement, read it back to verify what the zone actually serves. The dangerous window between delete and create becomes seconds inside one execution instead of an unlogged console session.
Every resource group ships a List task with fetchType control: NONE for counts, FETCH for rows, STORE for ion files ready for DuckDB. A weekly parallel sweep counts droplets, volumes, databases, load balancers, DOKS clusters, firewalls, and domains, and posts one Slack digest before the invoice surprises anyone.
Patterns platform teams run in production today. Each one shows the flow end to end, with the real plugin classes in play.
The flow creates a droplet named after the execution ID, polls droplet.Get until DigitalOcean reports it active, locks it down with a per-run firewall, and runs the workload injected via cloud-init. Teardown of both firewall and droplet lives in a finally block, so no failure path leaks a paid VM.
The droplet exists only for the duration of one execution, sized per run through flow inputs.
LoopUntil polls every 10 seconds and fails the run if the droplet is not active within 10 minutes.
The finally block deletes firewall and droplet on success, crash, or timeout, skipping unused resources.
Two schedules pass different size slugs into one flow. It refuses to act unless database.Get reports the cluster online, submits database.Resize, then polls until the migration completes. The Slack confirmation means the change actually landed, not that an API call returned 202.
The sizing policy lives in version-controlled YAML with per-trigger inputs, not in a console reminder.
A Fail task stops the run when the cluster is already resizing, instead of stacking a second operation.
Polling continues until status returns online, with a ceiling that fails loudly, not silently.
A nightly schedule lists the whole fleet, walks it with a concurrency limit of 2 to respect API rate limits, and fires droplet.Action with SNAPSHOT for each machine carrying the tag. Snapshot names embed the date, so retention tooling sorts and expires them without guesswork.
New machines join the backup policy by adding the auto-backup tag. No server inventory file goes stale.
Each snapshot is its own task run inside the ForEach; one API error leaves every other backup untouched.
A distinct alert fires when listing or snapshotting fails, separate from the nightly summary heartbeat.
droplet.Trigger fires an execution for every new droplet on the account. A naming rule checks the first hyphen-separated segment against approved prefixes; violations get POWER_OFF rather than deletion, so a mis-named machine keeps its disk and can be revived after review.
Console click, doctl, or API script: any new droplet is seen within one polling interval.
Power off preserves the disk. A false positive costs a rename and a restart, not a restore from backup.
Compliant droplets get a log line, so history records everything that ever joined the account.
The flow lists the zone, deletes every existing A record for the hostname including accidental duplicates, creates the replacement pointing at the new environment with a 300 second TTL, then reads the record back so the Slack announcement quotes what DigitalOcean actually serves. A failed half-applied cutover pages the team immediately.
Zones accumulate duplicate A records from manual edits; the delete loop clears them all first.
The announcement quotes the record read back from the API, not the flow's inputs.
Re-run with the previous IP as input. The 300 second TTL bounds how long clients keep the old answer.
| Capability | DigitalOcean Console (manual) | doctl + cron on a droplet | Terraform DO provider | |
|---|---|---|---|---|
| Ephemeral droplet lifecycle with guaranteed teardown | finally block destroys droplet + firewall on every failure path | Manual delete you hope someone remembers | Script dies halfway, VM keeps billing | Made for standing infra, not per-job VMs |
| Wait until a droplet is actually active | LoopUntil polls droplet.Get, fails loudly on timeout | Watch the spinner | Hand-rolled sleep loop | Provider waits, but only inside apply |
| Scheduled managed database resize | Schedule + Resize + poll until online, per-trigger sizes | One-off manual action | doctl resize in cron, no completion check | Plan/apply per change, no calendar |
| Tag-filtered fleet snapshots with dated names | List + ForEach + Action SNAPSHOT, per-droplet isolation | Weekly built-in backups, fixed cadence, no tag filter | Custom script with pagination and retries | Not an operations tool |
| React when a droplet appears on the account | droplet.Trigger fires one execution per new machine | Nothing reacts | Diff two doctl outputs yourself | Detects drift only when you run plan |
| DNS cutover with duplicate cleanup and verification | record List + Delete + Create + Get in one execution | Console edits with an unbounded delete-create window | Sequential doctl calls, no rollback story | Records as state, cutover as a config edit |
| Whole-account inventory on a schedule | Seven parallel List tasks, totals-only mode, Slack digest | Seven console pages | Custom script + Slack plumbing | State file lists managed resources only |
| Chain DigitalOcean with Ansible, Kubernetes, Postgres, Slack | Outputs flow between plugins in one execution history | No chain | Custom glue per pair | Provider per tool, no runtime semantics |
| Self-hosted, air-gapped, OSS edition | Self-hosted by default, OSS edition free | SaaS console | Self-hosted scripts, self-owned toil | Self-hosted or HCP |
Find answers to your questions right here, and don't hesitate to Contact Us if you couldn't find what you're looking for.
Automate infrastructure workflows in Kestra
How Kestra orchestrates cloud providers, Ansible, Terraform, and Kubernetes in one declarative flow.
Pair DigitalOcean with Ansible
Provision the droplet with Kestra, configure it with Ansible, tear both down from the same execution.
DigitalOcean plugin reference
Every task and trigger (droplet.*, database.*, volume.*, loadbalancer.*, kubernetes.*, firewall.*, domain.*) with copy-ready YAML and full property reference.
Ephemeral droplets with guaranteed teardown, business-hours database scaling, tag-driven snapshot policies, governance reactions to every new machine, and verified DNS cutovers. Open source, self-hosted, event-driven.
