From 09e39ca6aaa6267393f695658720b1efde2a9ff6 Mon Sep 17 00:00:00 2001 From: Walter Behmann Date: Tue, 7 Jul 2026 17:09:21 +0200 Subject: [PATCH] Document dedicated master nodes on the Services page Explain what dedicated master nodes are, when they help, how they are chosen at deploy time (odd quorum, fixed size, spread across AZs), and how they are billed. Link the Custom-plan feature bullet to the new section. --- CHANGES.rst | 1 + docs/reference/services.md | 43 +++++++++++++++++++++++++++++++++++++- 2 files changed, 43 insertions(+), 1 deletion(-) diff --git a/CHANGES.rst b/CHANGES.rst index 1a0e08c..24b3231 100644 --- a/CHANGES.rst +++ b/CHANGES.rst @@ -6,6 +6,7 @@ Changes Unreleased ========== +- Document dedicated master nodes on the Services page - Consolidate ``cloud-reference``, ``cloud-howtos``, and ``cloud-tutorials`` - Adjust links to accompany renaming to ``cloud-docs`` - Adjust intersphinx references to accompany consolidation into single repository diff --git a/docs/reference/services.md b/docs/reference/services.md index a605e72..9a066b4 100644 --- a/docs/reference/services.md +++ b/docs/reference/services.md @@ -50,7 +50,7 @@ _large production workloads_ - Any cluster size - Custom compute options -- Dedicated master nodes +- {ref}`Dedicated master nodes ` - Unlimited Storage - Custom Backups - Premium Support @@ -154,6 +154,47 @@ three (e.g., 3, 6, 9, 12, etc.) provides optimal distribution across zones, it is not strictly required for high availability. +(dedicated-master-nodes)= +## Dedicated master nodes + +By default, every node in a cluster is master-eligible: on top of storing and +querying data, any node can be elected to manage the cluster state—keeping +track of which nodes are members, where shards live, and coordinating changes +to that metadata. On smaller clusters this shared arrangement works fine. On +large or busy clusters it can work against you, because the elected master +competes for CPU and memory with the same query and ingest traffic it is meant +to coordinate—the data node under the most load is also the one holding the +cluster together. + +Dedicated master nodes split those two jobs apart. The cluster runs a small, +separate pool of nodes that only manage the cluster state; they hold no data +and answer no queries, so coordination stays responsive no matter how hard the +data nodes are working. In practice this means steadier cluster management and +cleaner failover on larger deployments. + +**When to use them**
+Consider dedicated master nodes once a cluster grows past a handful of data +nodes, sustains heavy ingest or query load, or has strict availability +requirements. Smaller clusters don't need them—the data nodes take on the +master role without any trouble. + +**How they work**
+- You opt into dedicated master nodes when deploying a cluster. They are + available on the larger dedicated sizes (CR3 and up). +- The pool always runs an odd number of nodes—three by default, optionally + five—so a master can always be elected by a majority (quorum). An even number + adds cost without improving fault tolerance, so it isn't offered. +- Master nodes are distributed across availability zones like data nodes, and + each is deployed with its own storage. Their size and count are set when the + cluster is created and stay fixed when you scale the data nodes up or down. + +**Billing**
+Dedicated masters are billed as additional nodes: compute for each master while +the cluster is running, plus their storage. Suspending a cluster pauses master +compute along with the data nodes; master storage keeps being billed, the same +way data storage does. See {ref}`billing ` for details. + + ## Custom For organizations with specialized requirements that go beyond the Shared and