Elastic Cloud / managed offerings

Elastic Cloud is Elastic’s hosted, managed Elasticsearch + Kibana service that runs deployments on AWS/GCP/Azure and automates provisioning, upgrades, scaling, snapshots, and security so teams skip cluster operations.

Why it matters

Running Elasticsearch well is real operational work — capacity planning, rolling upgrades, snapshot schedules, TLS, node failure recovery. Managed offerings absorb that toil, which is why most new adopters start there. Knowing the landscape (Elastic Cloud vs AWS’s own service vs self-managed) also prevents a costly trap: the AWS “OpenSearch” fork is a different product and not API-compatible with current Elasticsearch.

How it works

You declare a deployment topology by hardware tiers; the platform maps that onto node roles and an ILM data-tier layout and manages the lifecycle.

  • Deployment building blocks — sized hot/warm/cold/frozen data tiers, dedicated masters, ML and ingest nodes; you pick RAM/zone counts, not individual VMs.
  • Frozen tier + searchable snapshots — the cold/frozen tiers back data with object storage, so a snapshot is the searchable copy; you pay storage, not hot hardware, for old data.
  • Managed upgrades — version bumps are orchestrated rolling restarts with one click, with snapshots taken first.
  • Autoscaling — tiers can grow on capacity/storage triggers tied to ILM policies.
OfferingWho runs itCompatibility
Elastic Cloud (ESS)Elasticcurrent Elasticsearch + full X-Pack
Elastic Cloud Enterprise / on K8syou, via Elastic’s operatorsame product, your infra
AWS OpenSearch ServiceAWSforked from 7.10; not Elasticsearch

Example

A logging deployment spanning tiers under one ILM policy:

hot   (NVMe, 64GB RAM x3 zones)  -> active indexing, last 2 days
warm  (HDD,  cheaper x2)         -> 2–14 days, force-merged, read-only
cold  (searchable snapshot)      -> 14–90 days, data in object store
frozen (searchable snapshot)     -> 90d–1y, fetched on demand

ILM rolls each data stream index down the tiers automatically; the frozen tier keeps a year queryable at a fraction of hot cost.

Pitfalls

  • OpenSearch is a hard fork — code, APIs, and clients diverged after Elasticsearch 7.10’s license change; do not assume new ES features, the _sql/Painless behavior, or official clients work against it.
  • Cross-zone bandwidth and egress — multi-AZ replication and snapshot egress to other clouds add cost that the sticker price hides; keep workloads and storage in-region.
  • Managed != tuning-free — the platform won’t fix oversharding, bad mappings, or runaway aggregations; you still own data modeling and query hygiene.
  • Version cadence is enforced — managed services drop very old versions on a schedule; plan upgrades rather than pinning indefinitely.

See also