Running Drone CI?

Drone pipelines failing on secrets or steps? We unstick it.

Drone keeps its appeal because every step is a container and nothing else, which is also the reason it falls over in the exact same place every time: a secret that is not available on a pull request, a service container that never became ready, or a runner that has quietly lost the Docker socket. We take the pipeline apart from the clone step down to the last plugin and bring it back to a known good state.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

drone-runner.log

still broken

What it is

Drone CI at a glance

Drone is a container-native continuous integration system started by Brad Rydzewski in 2014 and now owned by Harness. Every step runs in its own container, pipelines are declared in a single .drone.yml file at the root of the repo, and the server speaks to GitHub, GitLab, Gitea and Bitbucket through standard webhooks.

The architecture splits cleanly into a server and one or more runners, with runners for Docker, Kubernetes, Exec and SSH. The simplicity of the file format is the headline feature: anything you can put in a container you can run as a step, and anything that needs state becomes a service container. There is still a community edition, though the project has shifted gravity toward Harness CI since the acquisition.

Type
Container-native continuous integration
First released
2014
License
Apache 2.0 community edition, Harness enterprise above that
Maintained by
Harness, Inc. and the Drone community
Typically used for
Container pipelines, self-hosted CI, lightweight runners next to Gitea or GitLab

The honest version

Drone CI pros and cons

No technology is all upside. This is where Drone CI earns its place and where it tends to cause trouble.

Where Drone CI is strong

  • One file, one mental modelEvery step is a container, every service is a container, and the YAML is short enough that a new developer reads the whole pipeline on their first day.
  • Self-hostable end to endServer and runners both run inside your own infrastructure, which pairs well with self-hosted Git providers like Gitea and on-prem GitLab.
  • Lightweight to operateThe server is a single Go binary with a small SQLite or Postgres backing store, which is a welcome change after wrestling with heavier CI control planes.
  • Kubernetes runner fits modern platformsPipelines execute as pods, so the same cluster that runs the apps can run the build, and resource requests are applied per step.
  • Plugin model keeps steps portableDocker, Slack, S3, GCS and dozens of community plugins behave the same way across pipelines because they are just images with a documented interface.
  • Trigger filters are expressiveEvent, branch, path and ref filters in the YAML let you avoid running the deploy pipeline on every pull request without a wall of shell.

Where Drone CI bites

  • Secret scoping bites forksSecrets are gated on event and repo by default, which is correct, but the error at build time only says the secret is missing, which sends teams off chasing the wrong cause.
  • Service containers are flaky to orchestrateDrone does not wait on a port to open, so Postgres or Redis service containers need a wait-for script or the first step races and fails.
  • Observability is sparseNo analytics, no flaky test detection and limited log search. Teams end up shipping build logs to their own stack to get any real insight.
  • The community edition is quietly behindFeature work has moved toward Harness CI, so new capabilities in that product take a while to land, or do not land, in the open-source server.
  • Runners can lose the Docker socketA Docker daemon restart, SELinux policy, or a container that remounted the socket can leave the runner healthy on paper but unable to start any step.
  • YAML schema is strict and version-specificA field added in a newer runner will be reported as unknown on an older one, and the error does not tell you which version you are actually running.

Sound familiar?

Common Drone CI problems we fix

The Drone CI problems that bring people to us, and what you have once each one is fixed.

secret not found: DOCKER_PASSWORD

Secrets refuse to appear in pull request builds

A secret that is not marked as available on pull request events, a repo that inherited org-level restrictions, or a trigger block that silently filtered the event the secret needs to come through on.

Secrets reach only the builds they should

step service postgres: port 5432

Service containers time out

Drone starts the service in parallel with the first step and never waits for the port to open. Tests hit the socket before the database has finished initdb and die.

Services ready before the step runs

exit code 137 (killed)

Steps are killed by the kernel

A container that exceeded the memory limit on the runner, or a Kubernetes pod with no memory request landing on a node that is already full.

Steps sized against real usage

cannot pull image: unauthorized

Plugin images will not pull

A private registry that was never wired into the runner, a credential in the server database that has expired, or a pipeline using a tag that was deleted upstream.

Pulls authenticated and pinned

Cannot connect to the Docker daemon

The runner cannot talk to Docker

A host daemon restart that broke the socket bind, an SELinux rule, or a runner container that was rebuilt without the socket volume being wired back in.

Runner healthy, socket mounted

unknown field 'depends-on'

Pipeline fails on a field the docs mention

A YAML feature added in a newer runner version with the old runner still in service, or an exec runner mistakenly handling a Kubernetes-only directive.

Runner and schema versions lined up

cache plugin: s3: AccessDenied

Build cache uploads fail

An IAM role that lost its bucket permission, a bucket region change, or a cache prefix that collides with another repo and now hits object-level ACL rules.

Cache reads and writes clean

signature verification failed

Webhooks silently stop triggering

A shared secret rotated on the Git provider but not on Drone, a reverse proxy that strips the signature header, or a webhook URL that was moved and never updated on the provider.

Webhooks arrive and authenticate

What we do

Drone CI help, from runner rescue to pipeline work

Reach out for one broken step, or let us take on the whole Drone install.

Rescue
Pipelines wedged, runners gone, or secrets not reaching the right steps. We get builds green first and document the cause second.
Runner fleet
Docker, Kubernetes and exec runners sized against your workload, including a sensible split between Linux amd64 and arm64.
Pipeline rewrites
Service container readiness, cache keys, path and event triggers, and matrix steps written so the file is short enough to read in one sitting.
Secrets and signing
Secret events scoped correctly, signed commits enforced on protected branches, and plugin inputs that never leak into the step logs.
Migration to or from Drone
Moves to or from Jenkins, GitLab CI or Harness, with pipelines translated step for step rather than dumped through an automated converter.
Managed Tech
Server upgrades, runner patching, a dashboard watching queue depth, and a team to ring when a release train derails on Friday afternoon.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running Drone CI alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

Drone CI questions

Something else? Call+1 737 727 7327

Why does my Drone secret say not found on pull requests?

Secrets default to not being available on pull_request events, because otherwise anyone opening a PR could read them. If the step needs the secret, mark it available on that event in the secrets settings for the repo and confirm the trigger block on the step allows the same event. The error only says missing, not restricted, which sends people hunting the wrong file.

How do I make service containers wait until they are ready?

Drone starts services alongside the first step rather than before it, so put a short wait-for-port script at the top of the test step. A loop around nc -z host port with a sensible timeout costs almost nothing and removes an entire class of flaky builds.

Why do my Drone pipelines get OOM killed?

Container memory limits on the runner, or on the Kubernetes node, are lower than the job needs at peak. Set resource requests on the Kubernetes runner or raise the limit on the Docker runner, and in test pipelines check whether the language runtime is holding onto memory it does not need.

Can we self-host Drone next to Gitea or GitLab?

Yes, that is one of the strongest cases for Drone. The server and runners both run on your own infrastructure, the OAuth hook points at Gitea or GitLab, and the whole thing stays inside your network perimeter. We have set up this pattern a number of times and it is stable once it is in place.

Is Drone still a good choice given Harness CI exists?

If the community edition does what you need and you want a small, self-hosted CI next to the rest of your stack, Drone is still a reasonable call. If you need the newer analytics, governance and test insights, Harness CI is now where that work lives, and we can advise on whether moving across pays for itself.

How much does Drone CI help cost?

Our estimates are free of charge. Send us the pipeline and a description of what is going wrong, and a quote lands before any work starts.

Free estimate

Drone CI giving you trouble?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.