Apache Airflow support, 24/7

Apache Airflow Support for Kubernetes, Celery, MWAA, Cloud Composer and Astronomer

Last reviewed by Tyler Eastridge

AceMQ supports Apache Airflow 24/7 on Kubernetes, Celery, Amazon MWAA, Google Cloud Composer and Astronomer. Most calls start with one of three problems: a scheduler that stops scheduling with no error, DAG files that take so long to parse that every run starts late, or tasks that sit in queued while executor slots are held by dead tasks.

24/7 across Americas, EMEA and APAC, with response targets from 15 minutes, by plan, and a written root cause analysis on every P1.

Support quotes answered within 1 hour, by a senior engineer.

Planning a build, upgrade or migration instead? See AceMQ Apache Airflow consulting and services.

160+
Customers supported
10+
Years in production support
40+
Sev 1 incidents handled a year
24/7
Americas, EMEA and APAC
1hour
Quote reply

Customers

  • Microsoft
  • Henry Schein
  • Enbridge
  • Gojek

Partners

  • Broadcom
  • Redis
  • VMware Tanzu
  • OSSeva

Incident Triage

Airflow problems we fix every week

Silent schedulers, late DAGs, tasks stuck in queued and bloated XCom: each symptom matched with the setting, log or table we open first.

Common Apache Airflow production problems and how AceMQ resolves them
Symptom you're seeingWhat we check firstTypical resolution
DAGs stop scheduling, no error in the UIScheduler heartbeat; max_active_runs saturation on the DAG; executor slots held by zombie tasks left by a crashed workerReconcile stuck task instances against live workers, release the slots, align zombie thresholds with real task runtimes
Everything starts late, across every DAGPer-file parse time in the DAG processor logs; top-level API calls, Variable or secrets lookups and heavy imports run on every parseMove work from module scope into task callables; cache metadata for dynamic DAG generation
A task stays queued and never startsThe task's queue name against the queues workers listen on; pool slots; parallelism and max_active_tasksAlign queue names, correct pool and parallelism settings
Metadata database slow, XCom table hugeWhat flows through XCom; dataframes and large payloads stored there instead of a reference to object storagePass references, run airflow db clean on a retention rule
Metadata database connections exhaustedPool sizes on scheduler, webserver and workers against DAG concurrency; SQLAlchemy pool_sizeSize pools against real parallelism, add a connection pooler
Workers killed and task state lost (Celery or Kubernetes executor)Pod memory limits against task memory; evictions under resource pressureRight-size pods, set per-task-class resource requests
A DAG ran at the wrong time, or ran twiceschedule, timezone, start_date and catchup; timezone-naive schedules plus catchup=True and a backfillMake schedules timezone-aware, set catchup deliberately
Sensors hold worker slots for hoursBlocking sensors in reschedule or poke modeConvert to deferrable operators

P1 closure includes a note on the DAG, pool or worker setting that caused the stall, and what we changed to stop a repeat.

Scheduler stopped and a critical DAG chain not running? Ask for a quote and we answer within an hour.

Contact Us

Deployment coverage

Which Airflow deployments does AceMQ support, and who covers what?

DeploymentWhat AceMQ coversWhat the vendor coversExcluded
Self-managed Airflow on Kubernetes (official Helm chart)24/7 incident response, written RCA on P1s, scheduler and executor tuning, metadata database care, upgradesThe Airflow community offers help on GitHub and Slack with no response commitment (airflow.apache.org, read 17 Nov 2026)Your DAG business logic is reviewed, not rewritten
Airflow on VMs or DockerThe sameThe sameHost faults
CeleryExecutor and KubernetesExecutor deploymentsQueue routing, autoscaling, pod templates, executor migrationThe sameBroker faults for Celery's message broker are covered at the configuration level
Amazon MWAADAG design, pool and slot behaviour, dependency conflicts, upgrade planningAWS runs the service and the Airflow versions it offersAWS platform faults, settings MWAA does not expose
Google Cloud ComposerThe same layerGoogle runs the servicePlatform faults
Astronomer (Astro)The same layer; the work between your DAGs and the platformAstronomer runs its platform under its plan terms (astronomer.io/pricing, read 17 Nov 2026)Astronomer platform faults
Managed Airflow in your own account (optional extension)AceMQ takes monitoring, on-call, patching, upgrades and capacity on your Kubernetes or VMs; you keep the DAGsn/aDAG code, which your teams own

Support options

Who provides enterprise support for Apache Airflow? Vendor and independent support compared

Apache Airflow has no support contract from the project. Enterprise support comes from managed-service vendors for their own platforms, such as Astronomer, Amazon and Google, and from independent providers such as AceMQ for Airflow you run yourself or want kept where it already runs.

Astronomer / Cloud managed services (MWAA, Composer)
AceMQ independent support
What it coversAstronomer: Astronomer's platform (Astro) and Astro Private CloudCloud managed services (MWAA, Composer): The provider's managed Airflow environment
What it coversAirflow on any infrastructure, and the DAG layer on managed services
Airflow versionsAstronomer: Per Astronomer; the pricing page lists noneCloud managed services (MWAA, Composer): Per provider
Airflow versionsThe version you run
DAG authoring and executor tuningAstronomer: Largely yoursCloud managed services (MWAA, Composer): Yours
DAG authoring and executor tuningPart of the work: we read your DAGs when the problem lives there
Existing deploymentAstronomer: A move onto Astro is a migrationCloud managed services (MWAA, Composer): A move onto the service is a migration
Existing deploymentStays where it runs, on versions and configuration you choose
PriceAstronomer: Astronomer plansCloud managed services (MWAA, Composer): Cloud usage
Price$24,000 to $72,000 a year (USD), set by SLA, scope and deployment. Quote reply within 1 hour

Short note: a managed service removes some operations, not DAG design, executor slots or metadata database pressure, which cause most incidents. Teams on Astro, MWAA or Composer often add independent cover for that layer.

What's Included

Everything in your Apache Airflow support contract

Scheduler and executor tuning, DAG parse time, the metadata database, provider pins and upgrades are covered on managed Airflow and on clusters you run.

  • 01

    Emergency Incident Response

    The scheduler stops scheduling, or a business-critical DAG chain is stuck. With response targets from 15 minutes, by plan, a senior engineer joins a live bridge with direct access to diagnose — not a ticket acknowledgement.

  • 02

    Root Cause Analysis

    Every P1 closes with a written RCA: what failed, why, the fix applied, and the specific DAG, config, or infrastructure change that prevents recurrence.

  • 03

    Scheduler & Executor Tuning

    max_active_runs, pool sizing, executor slot allocation, and DAG parse time reduction tuned against your actual DAG count and task concurrency.

  • 04

    Upgrade & Compatibility Advisory

    Guidance on major version upgrades, provider package compatibility, and custom operator migration — including the DAG serialization changes that break naive 1.x-to-2.x upgrades.

  • 05

    Capacity & Metadata DB Reviews

    Quarterly reviews of metadata database connection load, XCom growth, and worker resource headroom so scheduler performance doesn't degrade as DAG count grows.

  • 06

    Migration & Architecture Support

    Executor migration (Celery to Kubernetes or vice versa), consolidating legacy cron and Pentaho-style batch jobs into Airflow, and DAG design review for teams scaling past a few hundred DAGs.

Why AceMQ

What teams get from a dedicated Airflow engineer

  • Named Engineers, Zero Cold Start

    The same senior engineers stay on your account. They know your DAG topology, your executor setup, and which pipelines are fragile — so a P1 starts with diagnosis, not twenty minutes of explaining your environment.

  • No Tier-1 Triage Layer

    You reach a senior Airflow engineer directly by phone, email, or Slack. No help desk collecting information to pass along, no escalation approval process standing between you and someone who can fix it.

  • Orchestration and the Systems Underneath

    Most Airflow incidents are metadata database, executor infrastructure, or DAG design problems wearing an Airflow error. We diagnose across the whole stack, not just the scheduler UI.

  • Genuine Follow-the-Sun Coverage

    Round-the-clock coverage for customers in 26+ countries. Your overnight batch failure is someone's mid-afternoon — no overnight skeleton crew, no waiting for a region to wake up.

  • Proactive, Not Just Reactive

    Quarterly health checks on DAG parse time and metadata DB load, plus shared intelligence across our support base. When a provider package bug surfaces on one customer's DAGs, every affected customer hears about it before it reaches their production.

  • We Read Your DAGs, Not Just Your Logs

    Airflow failures are frequently caused by DAG authoring patterns — top-level API calls, unbounded XCom, missing idempotency — that logs alone won't reveal. We review the actual DAG code when that's where the problem lives.

Airflow platforms in scope

Amazon MWAA, Cloud Composer, Astronomer, self-managed Airflow on Kubernetes, Celery and Kubernetes executors, and networks closed to the outside.

  • Amazon MWAA
  • Google Cloud Composer
  • Astronomer
  • Self-managed on Kubernetes
  • AWS (EC2, EKS)
  • Google Cloud (GKE)
  • Microsoft Azure (AKS)
  • CeleryExecutor deployments
  • KubernetesExecutor deployments
  • Air-gapped / no outbound access
  • Airflow 2.x

Version coverage

Airflow versions we support

Airflow versions we support
Airflow lineWhat the Airflow project lists (airflow.apache.org supported versions, read 17 Nov 2026)What AceMQ does on it
Airflow 3.x3.4.0 current patch, state "Maintenance", first release 22 Apr 202524/7 support and the target for upgrades
Airflow 2.x2.11.2 is the last 2.x release listed; first release 17 Dec 202024/7 support and a planned upgrade to Airflow 3
Airflow 1.101.10.1524/7 support and a staged path through 2.x

Airflow 3 is an architectural release: tasks no longer talk to the metadata database directly, the REST API changed, SubDAGs and other deprecated features were removed, datasets became assets and a newer Python is required. The project recommends moving to the last 2.x release first and clearing deprecation warnings there, and it ships a linter that flags incompatible DAG code, which turns the upgrade into a measurable backlog. Estates on 1.10 go through 2.x on the way.

In practice

What an Airflow 3 migration looks like when we run it

Inventory. Airflow, Python and provider package versions; every DAG using SubDAGs, SLAs, direct metadata database access or datasets.

Lint. Run the project linter across the DAG repository and turn the output into a ranked backlog per team.

Environment build. A parallel Airflow 3 environment with pinned providers and constraint files.

Parallel run. The same schedules run against production data in a safe target.

Cutover with rollback. DAG groups move in waves; the previous environment stays as the rollback until each wave is verified.

Your teams keep owning DAG code. We handle the platform, the provider alignment and the evidence.

Consulting

Apache Airflow consulting and professional services

Apache Airflow consulting and professional services
DeliverableWhat you receive
Platform assessmentScheduler, executor and metadata database configuration against measured load; DAG authoring audit (idempotency, retry safety, top-level code); secrets and connection review; upgrade readiness
Executor migrationTask duration and concurrency profile, Celery or Kubernetes recommendation including hybrid designs, staged DAG-group migration with rollback
DAG performance reviewPer-file parse timing, ranked offenders, parse-time refactors, tuned min_file_process_interval
Metadata database careRetention rule with airflow db clean, growth monitoring, backup and restore tests
Upgrade planAirflow 2 to 3, provider pinning, rehearsal on a copy
Managed operations in your accountRunbooks, monitoring, on-call and upgrade calendar; see managed Airflow

Escalation Path

Your first hour of an Apache Airflow outage

A scheduler that has stopped is answered ahead of a question about DAG layout. Your plan sets each response clock below.

  1. From 15 min (by plan)first responseEmergencyProduction down: scheduler not scheduling, a business-critical DAG chain stuck, or the metadata database unreachable
  2. 1 hourfirst responseCriticalSevere degradation: every DAG starting late from parse time, executor slots held by zombie tasks, workers killed and task state lost
  3. 4 hoursfirst responseHighTasks stuck in queued, connection exhaustion on the metadata database, DAGs running twice after a timezone or catchup change
  4. Next Dayfirst responseStandardQuestions on pools, sensor modes, XCom usage, provider upgrades and migrating from Airflow 2 to Airflow 3

Plan and severity set the clock: a halted scheduler and a retry-policy question are never in one queue.

Want an Airflow engineer who knows your DAGs and executor setup? Request a quote or message us.

Contact Us

Pricing

What does AceMQ Apache Airflow support cost?

Airflow support is typically $24,000–72,000 a year. The number of environments, DAG and task volume, your executor, and managed or self-managed hosting move the quote, along with the response plan.

Prices are in US dollars (USD). Response targets start at 15 minutes by plan, 24/7, and quote requests are answered within 1 hour. Covering several pipeline tools in one agreement can earn a multi-technology discount.

Support In Practice

Airflow problems we have fixed

Apache Airflow support work from AceMQ. Each card links to the full write-up: the Airflow estate, what AceMQ diagnosed or changed, and the result.

All use cases

Support options

Four tiers of Apache Airflow support.

AceMQ offers Apache Airflow support in four tiers, each handing more of the day-to-day work to AceMQ: 24/7 support and incident management for platforms your team runs, proactive monitoring that finds parse delays and slot exhaustion first, managed Airflow on your own infrastructure, and fully managed Airflow on AceMQ infrastructure. Airflow 3 and Airflow 2 are covered, including MWAA, Cloud Composer and Astronomer environments.

  • Tier 1

    Apache Airflow 24/7 Support & Incident Management

    Your data engineers own DAGs, we take escalations

    Best for: Teams that write and run Airflow themselves and need a scheduler and executor specialist for the worst days.

  • Tier 2

    Apache Airflow 24/7 Support with Proactive Monitoring

    We profile DAG parsing, pools and metadata growth

    Best for: Teams whose DAG count keeps growing and who want parse time and database load checked before the scheduler slows.

  • Tier 3

    Managed Apache Airflow on Your Infrastructure

    We operate Airflow in your account

    Best for: Organizations that keep Airflow on their own Kubernetes or cloud account while we handle monitoring, patching and upgrades.

  • Tier 4

    Fully Managed Apache Airflow on AceMQ Infrastructure

    We run your DAG scheduler and workers as a service

    Best for: Data teams that want to write DAGs and leave the platform, on-call and capacity to someone else.

FAQ

Apache Airflow support questions

Still have questions about Apache Airflow? Email an Expert

Who provides enterprise support for Apache Airflow?

Astronomer supports its own platform, and AWS and Google support MWAA and Cloud Composer. The Airflow project offers community help on GitHub and Slack with no response commitment. Independent providers such as AceMQ support Airflow on any infrastructure, 24/7, with response targets from 15 minutes, by plan.

What does 24/7 Airflow support cover with AceMQ?

Round-the-clock response from 15 minutes on top plans, a written root cause after each P1, plus tuning, upgrade advice and capacity reviews. Example Sev 1: the scheduler has stopped with no UI error. An engineer checks the scheduler heartbeat, max_active_runs and zombie-held executor slots first. P2 is answered in 1 hour.

Why did my scheduler stop scheduling with no error?

Three causes cover most cases: scheduler heartbeat failure, max_active_runs saturation on one DAG, and executor slots held by zombie tasks after a worker crash. We check all three in the first diagnostic pass.

Why are all my DAGs starting late?

Probably parse time. Heavy top-level code in one DAG file, such as API calls or expensive imports, runs on every parse and delays scheduling for the whole instance. We rank files by parse cost and refactor the offenders.

Why is a task stuck in queued?

Usually a queue name that no worker listens on, a pool with no free slots, or executor parallelism exhausted by dead tasks. We align queue, pool and worker configuration.

Do you support Amazon MWAA, Google Cloud Composer and Astronomer?

Yes, alongside self-managed Airflow. Managed services remove some operations but not DAG design, executor slot exhaustion or metadata database pressure.

Can AceMQ run Airflow for us?

Yes, as an extension of support. AceMQ runs the platform in your own account: monitoring, on-call, patching, upgrades and capacity, while your team writes DAGs. See managed Airflow.

Can you help us migrate to Airflow 3?

Yes. We inventory removed features, run the project's linter, build a parallel environment, run production schedules in parallel and cut over in waves with a rollback.

Do you support custom operators and providers after an upgrade?

Yes. Provider package pins and DAG serialization rules are the usual breakages, and we validate and fix custom operators as part of upgrade support.

What does an Airflow health check include?

Scheduler loop time, DAG parse profile, executor and pool configuration, metadata database growth and retention, secrets handling, sensor usage, version and provider currency, and a ranked plan. Quotes are answered within 1 hour.

Who offers Apache Airflow consulting and professional services?

Astronomer and the cloud providers offer services around their platforms, and independent firms such as AceMQ consult on Airflow wherever it runs. Our consultants staff the 24/7 rota. See Apache Airflow consulting.

Can we hire AceMQ for one Airflow task and stop there?

Yes. A DAG parse-time audit, an executor migration or an Airflow 3 upgrade plan can be a stand-alone engagement. Teams with revenue-critical schedules usually add the annual agreement afterward, which brings 24/7 response and engineers who already know their DAGs.

How is Airflow support priced?

Most contracts land between $24,000 and $72,000 a year (USD). We scope by the number of environments, DAG count and task volume, Celery or Kubernetes executor, whether you use MWAA, Composer or Astronomer, and the response plan. Send those details and we reply with a quote within 1 hour.

Let's talk

Make the Morning Report Late for the Right Reason

Send us your Airflow version, executor and DAG count. We reply to quote requests within 1 hour.

A senior Apache Airflow engineer, not a sales queueQuote requests answered within 1 hourAd hoc engagements or an annual partnership

Production incident right now?Email an Apache Airflow Architect