ElasticOps

Modern Elasticsearch cluster operations dashboard. Open-source successor to Cerebro.

Features

• Cluster Overview — Health metrics, node disk usage, active shard movements
• Shard Map — Pivot table (dynamic-depth rollup inferred from index names) and grid view (flat with shard chips)
• Index Explorer — Per-index analysis with optimization opportunity detection (over-sharded, under-sharded, segment fragmentation, and high deleted-doc ratio)
• Job Engine — Force merge, shard reduction, shard split (under-sharded → _split), deleted-doc expunge, reindex, and promote (atomic alias swap) with an approval workflow, plus operator-driven shard relocation and node drain, all run by an async background runner with live progress
• Node Management — Node health, storage breakdown, and node drain / undrain with live "shards left: N" progress straight from the Nodes view
• Settings Management — View/edit cluster settings with change tracking
• Multi-Cluster — Manage multiple Elasticsearch clusters from one dashboard; clusters can be marked read-only to block every write operation

Quick Start

git clone https://github.com/jugrajsingh/ElasticOps.git
cd ElasticOps

# Backend (from repo root)
uv sync --extra dev
uv run uvicorn backend.main:app --reload --port 4354

# Frontend (separate terminal)
cd frontend
npm install
npm run dev

Open http://localhost:5173 and add your Elasticsearch cluster connection.

Cluster Operations

ElasticOps runs operations against your cluster through an async background job runner that reports live progress and drives each job to a terminal state.

• Node drain / undrain — From the Nodes view, select a node and click Drain Node (an inline two-step confirm). Drain runs a pre-flight check first; if migrating the node's shards would be unsafe (e.g. too few remaining data nodes), it is refused and the reason is shown inline. Once draining, the node card shows live shards left: N progress and an Undrain button that cancels the drain and removes the allocation exclusion. Draining adds the node to cluster.routing.allocation.exclude._name (never clobbering nodes already excluded) and is fully reversible.
• Shard relocation — Move a single shard copy from one node to another via _cluster/reroute; the runner polls until the shard lands STARTED on the target.
• Shard split — Under-sharded indices are mapped to an ES _split job that creates <index>-split-<n> with the primary count raised to the smallest valid multiple (non-destructive; the source index is retained for you to verify before promoting/deleting).
• Deleted-doc expunge — Indices with a high deleted-doc ratio are mapped to a force-merge with only_expunge_deletes=true, reclaiming space without a full segment rewrite.
• Reindex — Copy an index into a new destination via the ES _reindex Tasks API (async, polled to completion); the source is left intact.
• Promote — Atomically swap an alias from a source index to a target (e.g. after a split/shrink) in a single _aliases call, with opt-in deletion of the old source.
• Rebalance advisory — From the Shard Map view, see suggested relocate moves that even out shard count across data nodes; each suggestion is executable as-is via the relocate job.
• Login rate limiting — Repeated failed logins from the same IP/account are throttled with a sliding-window lockout (in-process, returns HTTP 429), with successful logins clearing the counter.
• Read-only clusters — Mark a cluster read-only to block every write operation (drain, relocate, force merge, shard reduction, split, expunge, reindex, promote, settings edits). The backend enforces this rail and the UI hides the corresponding controls, so a read-only cluster cannot be mutated by accident.

Test Cluster

A local 3-node Elasticsearch cluster is provided for integration testing:

make -f Makefile.test test-cluster-up     # start 3 ES nodes (es01 on host port 9201)
ELASTICOPS_TEST_ES=http://localhost:9201 uv run pytest tests/integration -v
make -f Makefile.test test-cluster-down   # stop and remove the cluster + volumes

The integration tests exercise the real relocate / drain / force-merge / shrink executors against this cluster. They auto-skip when ELASTICOPS_TEST_ES is unset or the cluster is unreachable, so the default uv run pytest stays green without Docker.

Tech Stack

Layer	Technology
Backend	FastAPI + SQLAlchemy 2.0 (async)
Frontend	React 18 + Vite + Tailwind + shadcn/ui
Database	SQLite (default) or PostgreSQL
Tables	TanStack Table (virtual scroll)
State	TanStack Query (caching + auto-refresh)

Security

• On first run a random admin password is generated and printed to the logs; change it via Settings → Password (or POST /api/auth/me/password).
• The JWT signing secret and the credential-encryption (Fernet) key auto-generate into a gitignored .elasticops-secrets.json. For production set AUTH__JWT_SECRET and SECURITY__ENCRYPTION_KEY (env or local.env.yaml) so they are managed explicitly.
• Elasticsearch cluster credentials are encrypted at rest. Prefer least-privilege ES users.
• reduce_shards and split_shards jobs are non-destructive: they create a new copy (<index>-shrink-<n> / <index>-split-<n>) and leave the original index intact for you to verify before promoting/deleting.
• Login rate limiting is enforced app-side (sliding-window lockout, HTTP 429). (CSRF is N/A — auth uses Authorization: Bearer tokens, not cookies.)
• Known limitation — deactivating a cluster stops its status polling and blocks new ES operations (execute, execute-all, drain, promote, reindex, relocate, force merge) with HTTP 409, but a job already executing when the cluster is deactivated runs to completion; cancelling that job remains available regardless of the cluster's active state.
• Remaining roadmap: snapshot/restore, ILM policy management, Alembic migrations, and README screenshots.

License

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0) — see LICENSE.

Running a modified version over a network obligates you (under AGPL §13) to offer its complete source to users. A separate commercial license that lifts these copyleft/source-disclosure terms (e.g. to run a proprietary managed service) is available on request — contact jugrajskhalsa@gmail.com.