Emergency Incident Response
Coordinator down, queries hanging cluster-wide, or a connector that won't authenticate. With response targets from 15 minutes, by plan, a senior engineer joins a live bridge.
Query engine support
Last reviewed by Tyler Eastridge
AceMQ supports Starburst and Trino 24/7 across Americas, EMEA and APAC: Starburst Enterprise on Kubernetes or virtual machines, Starburst Galaxy, and open-source Trino. Most calls start with one of three problems: queries failing on per-node memory limits, a connector that will not push filters down, or Iceberg tables slowed by small files and long snapshot histories.
24/7 Starburst support with response targets from 15 minutes, by plan. Quote replies within 1 hour.
Planning a build, upgrade or migration instead? See AceMQ Starburst consulting and services.
Customers




Partners


Incident Triage
What we check first when a Trino or Starburst cluster misbehaves: the query plan, a catalog property, or the metadata tables of the table involved.
| Symptom you're seeing | What we check first |
|---|---|
| "Query exceeded per-node memory limit." | We read the failing query in system.runtime.queries and its stage details with EXPLAIN ANALYZE VERBOSE, then compare query.max-memory, query.max-memory-per-node and memory.heap-headroom-per-node with the worker heap. A join build side larger than a node is a plan problem (join order, distribution type, missing statistics); an undersized heap is a sizing problem. Spill (spill-enabled) and fault-tolerant execution (retry-policy=TASK with an exchange manager on object storage) are options with costs we explain first. |
| A filter does not push down and the network melts | We read the scan node in EXPLAIN for the constraint the connector accepted, then compare with the source's own query log. When the optimiser cannot push an expression, TABLE(system.query(query => '...')) sends a native query to a JDBC source, and a view or computed column can make the predicate sargable. |
| Iceberg queries slow down over weeks | We count files and snapshots from the "table$files" and "table$snapshots" metadata tables, then schedule ALTER TABLE ... EXECUTE optimize(file_size_threshold => ...), expire_snapshots(retention_threshold => ...) and remove_orphan_files. Writers that commit every few seconds need a compaction job, not a larger cluster. |
| Stale or slow metadata from Hive metastore or Glue | We look at hive.metastore-cache-ttl and hive.metastore-refresh-interval, call system.sync_partition_metadata for partitions added outside Trino, and check how long metastore calls take at the top of a slow query. |
| One dashboard starves the cluster | Resource groups (resource-groups.configuration-manager) with hardConcurrencyLimit, maxQueued and softMemoryLimit, and selectors on user or source, keep BI, ad hoc and scheduled work apart. We design them from a week of real query history. |
| Workers disappear mid-query | We check system.runtime.nodes for churn, spot or preemptible worker reclaim, and graceful shutdown settings (shutdown.grace-period) so a scale-down does not kill running tasks. |
| Catalogs differ between environments | A catalog properties file edited by hand on one coordinator is the usual cause. We move catalogs to version control and compare them with the running cluster. |
| Authentication and certificates | An expired TLS certificate or a changed OAuth2 or LDAP setting looks like a cluster outage to every client. We keep an inventory of certificates, authenticators and access control rules per cluster. |
Every closed incident includes the plan or property change, so the next engineer can see why it was made.
Deployment coverage
| Deployment | What AceMQ covers | What the vendor covers | Excluded |
|---|---|---|---|
| Starburst Enterprise (self-managed, Kubernetes or VMs) | Incident response, query and memory tuning, connectors and catalogs, resource groups, upgrade planning | Starburst supports the product and its defects under its Enterprise Support Policy (starburst.io/terms/enterprise-software-support, read 17 Nov 2026) | Product defects stay with Starburst; we work the escalation with you |
| Starburst Galaxy | Query design, catalogs, pushdown, table maintenance, cost drivers | Starburst runs the platform | Galaxy platform faults |
| Open-source Trino | Everything above, including upgrades, with no vendor behind it | No vendor contract; the project is community-run | Hardware and OS faults outside the agreed scope |
| Sources and storage (Iceberg, Delta, Hive, JDBC databases, object storage) | The Trino-side behaviour: connector settings, metadata, file layout, pushdown | The source vendor supports its own system | Source platform faults |
Support options
Starburst sells support for Starburst Enterprise and Galaxy. Open-source Trino has no vendor support contract. Independent support covers Trino you run yourself, and the query, data layout and source problems that sit outside a product vendor's scope.
What's Included
Coordinators, workers, catalogs and resource groups on Starburst Enterprise and Galaxy, plus the Trino underneath, under a single contract.
Coordinator down, queries hanging cluster-wide, or a connector that won't authenticate. With response targets from 15 minutes, by plan, a senior engineer joins a live bridge.
Every P1 closes with a written RCA: what failed, why, the fix applied, and the configuration change that prevents recurrence.
Pushdown validation, connector session property tuning, and worker memory sizing matched against your actual federated query patterns.
Resource group design so one heavy dashboard or ad hoc query can't starve every other workload sharing the cluster.
Configuration audits across multi-catalog federated setups to catch drift before it produces a query result nobody can explain.
Federated query architecture design, connector onboarding, and migration from single-engine query tools onto Starburst.
Why AceMQ
The same senior engineers stay on your account. They know your catalog topology and which connectors actually push down — so a P1 starts with diagnosis, not you re-explaining your setup.
You reach a senior Starburst engineer directly by phone, email, or Slack. No help desk, no escalation approval process standing between you and someone who can fix it.
Federated query problems live at the connector boundary, not just the Trino engine. We diagnose pushdown behavior across Iceberg, Delta, Hive, and JDBC connectors, not just the coordinator.
We support Starburst alongside Databricks and Snowflake as a connected stack, so a federated query problem that's really a source-system problem gets diagnosed end-to-end.
Round-the-clock coverage for customers in 26+ countries. Your 3am coordinator OOM is someone's mid-afternoon — no overnight skeleton crew.
Starburst problems are frequently metastore, network, or source-system problems wearing a query timeout. We diagnose across the whole path, not just the Trino layer.
Kubernetes and EC2 clusters, Hive metastore and Glue catalogs, Iceberg, Delta and Hudi tables, and networks with no route out to us.
Version coverage
| Line | Latest listed | Date | Note |
|---|---|---|---|
| Starburst Enterprise 483-e (short-term support release) | 483-e | 16 Oct 2026 | Starburst release notes, read 17 Nov 2026 |
| Starburst Enterprise 482-e (LTS) | 482-e | 11 Sep 2026 | Latest LTS line listed |
| Starburst Enterprise 480-e, 479-e, 477-e (LTS lines) | Listed | Various | Starburst states no support windows on the release page |
| Open-source Trino | 484 | 15 Oct 2026 | Trino release notes, read 17 Nov 2026 |
AceMQ supports all of these, and older versions you run today. Upgrades between releases are planned from your version, with a rehearsal against your real query log.
Escalation Path
A dashboard that dies on a memory error and a coordinator that will not start are not the same emergency, so each gets its own target.
Response clocks by plan. A cluster-wide outage starts the fastest timer, while sizing questions wait in the standard queue.
Pricing
Starburst support generally lands between $24,000 and $72,000 a year. What you pay follows the number of clusters and workers, how many catalogs and connectors you federate, whether you run Enterprise or Galaxy, and your response target.
Figures are in US dollars (USD). Response targets start at 15 minutes by plan, 24/7, and a quote request gets an answer within 1 hour.
Support In Practice
Starburst and Trino support and remediation work from AceMQ. Each card links to the full write-up: the federated estate, what AceMQ diagnosed or changed, and the result.
Public Sector Agency
Evaluating whether a federated query layer will actually work against a given set of source systems before the platform is committed to.
Global Payments Processor
Stopping worker crashes caused by unbounded joins, missing spill configuration, and memory limits that do not match the concurrency the cluster actually sees.
Energy Trading Firm
Named-engineer support for Starburst and Trino latency regressions, connector failures, and concurrency problems in production.
Enterprise Data Warehouse Team
Inventorying a Teradata estate, comparing Snowflake, Databricks and Trino against its workloads, and planning a pilot and a parallel run.
Global Logistics Operator
Making predicates, aggregates, and joins execute at the source connector instead of pulling full tables into Trino workers.
Support options
AceMQ offers Starburst support in four tiers, each handing more of the day-to-day work to AceMQ: 24/7 support and incident management for clusters your team runs, proactive monitoring that finds pushdown, memory and metastore problems first, managed Starburst Enterprise or Trino on your own infrastructure, and fully managed Trino on AceMQ infrastructure. Enterprise LTS releases and open-source Trino are covered.
Your engineers run Trino, we take escalations
Best for: Data platform teams that own the cluster and want a specialist for failed queries and coordinator incidents.
We tune queries, tables and groups with you
Best for: Teams watching Iceberg file counts, pushdown and memory limits slip and wanting them reviewed on a schedule.
We operate Starburst inside your environment
Best for: Organizations whose coordinators, workers and catalogs must stay in a private network or account.
We run the query layer as a service
Best for: Analytics groups that want federated SQL available without building a Trino operations practice.
One partner for Starburst
Our engineers also support the engines and stores around Starburst: Spark, ClickHouse, Druid and the object storage your tables live on.
Starburst supports its own Enterprise and Galaxy products, and open-source Trino has no vendor. AceMQ covers the cluster you operate, 24/7, with response from 15 minutes by plan. A Sev 1 with the coordinator down means an engineer on a bridge reading system.runtime.nodes first.
Usually a join build side or aggregation larger than one node can hold. The fix is a different join order or distribution type, better statistics, a lower query.max-memory-per-node ceiling for the whole cluster, or more headroom, in that order of preference.
One connector is not pushing predicates or aggregates down, so Trino pulls the whole table. Read the scan in EXPLAIN, then try a view, a different predicate or system.query pass-through.
Compact small files with optimize, expire snapshots on a schedule, remove orphan files, and keep partition design in line with how people filter.
Yes. We build them from real query history so BI, ad hoc and scheduled workloads each get a concurrency and memory share.
Yes, for everything you control on the query and data side: SQL, table layout, catalogs and access rules. Galaxy itself is run by Starburst, so platform-level faults go to them, and we help you gather the evidence for that case.
Replay a week of production queries against the new version, compare plans and results, upgrade workers and coordinator in a defined order, and keep the previous image ready for rollback.
A ranked report across memory and spill settings, failed and slow queries, connector pushdown, Iceberg and Hive table layout, resource groups, catalog files in version control, and certificate expiry. Each finding carries a severity and a proposed change.
Yes. A release upgrade replay, a resource group build or an Iceberg table cleanup can be bought as a separate engagement. If you later want 24/7 cover and fixed engineers who know your catalogs, the annual contract picks up from there.
A contract is usually $24,000 to $72,000 a year (USD). We look at cluster and worker counts, the catalogs and connectors you federate, Enterprise or Galaxy, how much unattended scheduled work runs, and the response level. Give us those and the quote arrives within 1 hour.
Let's talk
Send us your slowest query and its EXPLAIN ANALYZE output. We reply within 1 business hour with where the time is going.
Production incident right now?Email a Starburst Architect