InfluxDB

Editor’s note: This InfluxDB review is based on my hands-on test of InfluxDB 3 Core 3.12.1 on my own machine on October 22, 2026, plus a close look at the paid Enterprise and Cloud editions. I weighed pricing, ingestion and query performance, setup and security defaults, features, and how it compares with TimescaleDB, QuestDB and ClickHouse.

Quick verdict: I recommend InfluxDB for fast metric and IoT ingestion with recent-data dashboards, especially alongside Telegraf and Grafana. The trade-off: free Core caps a query at 432 Parquet files (about 72 hours per table) with no compaction or HA, so long history means Enterprise or Cloud.

Among real-time analytics databases, InfluxDB 3 runs SQL and InfluxQL on Apache DataFusion and Parquet, and I pushed about 427,000 points per second into it from a laptop with early write acknowledgments on.

In this review, I’ll take a closer look at InfluxDB’s pricing, features, and the Core limits that matter, so you can see exactly whether it fits your workload.

Key Takeaways:

• Core is free (MIT/Apache 2.0), a single binary that started in 9 ms
• ~427,000 points/s with no_sync, but ~10,000 points/s with durable writes
• Core refused my 60-day query after persistence: 432-file cap, no compaction
• Flux is gone in v3, so Flux tasks need rewriting
• Only Serverless publishes prices ($0.0025 per MB written); Enterprise is quote-only, except a free 2-core At-Home license

Pros and Cons

How Much Does InfluxDB Cost?

InfluxDB 3 Core is free, and Cloud Serverless is the only InfluxData edition with a public price list. The editions built for long history in production (Enterprise Commercial, Cloud Dedicated and Clustered) are all quote-only.

• InfluxDB 3 Core (free): for self-hosted, single-node workloads that query recent data; open source under MIT/Apache 2.0.
• Enterprise At-Home (free): for hobby and other non-commercial projects; never expires, includes every Enterprise feature, capped at 2 CPU cores per cluster.
• Enterprise Trial (free for 30 days): for evaluating Enterprise on up to 256 cores per cluster, no credit card needed.
• Enterprise Commercial (custom): for production; priced by CPU and sold in core batches of 8, 16, 32, 64 or 128.
• Cloud Serverless (Free plan, then usage-based): for managed, multi-tenant workloads; the Free plan allows 5 MB of writes per 5 minutes, 2 buckets and 30-day retention.
• Cloud Dedicated and Clustered (custom): for single-tenant managed hosting or self-managed Kubernetes; both are quote-only.

Serverless usage rates are $0.0025 per MB written, $0.012 per 100 queries, $0.002 per GB-hour stored (about $1.46 per GB-month) and $0.09 per GB of data out. InfluxData adds a $250 starting credit, with no credit card required.

Edition	Price	Hosting	Key limit
Core	Free	Self-managed, single node	432 Parquet files per query
Enterprise At-Home	Free	Self-managed	2 CPU cores per cluster; non-commercial use
Enterprise Trial	Free for 30 days	Self-managed	256 CPU cores per cluster
Enterprise Commercial	Custom, per CPU core	Self-managed	Core batches of 8 to 128
Serverless Free	Free	Managed, multi-tenant	5 MB writes per 5 minutes; 30-day retention; 2 buckets
Serverless Usage-Based	$0.0025/MB in, $0.012/100 queries, $0.002/GB-hour, $0.09/GB out	Managed, multi-tenant	50 MB per write request; 90-second query time
Cloud Dedicated	Custom	Managed, single-tenant	Priced by CPU/RAM and storage
Clustered	Custom	Self-managed Kubernetes	Not on the public price list

Amazon Timestream for InfluxDB 3 also runs Core or Enterprise on AWS, billed per node-hour with storage charged at a 200 GB minimum. Enterprise there adds a separate per-vCPU-hour license fee through AWS Marketplace, so price the full AWS bill, not just the instance.

Is InfluxDB Good Value for Money?

• Good fit: Core costs nothing for recent-data monitoring and edge nodes that don’t need HA, and you pay nothing until the file cap bites.
• Less convincing fit: long-history analytics, because Core hits the 432-file wall and Enterprise has no list price. Tiger Cloud starts at $30/month with published plans, so you’ll know your bill sooner there.
• Budget priority: At-Home suits homelabs. On Serverless, 1 GB (1,024 MB) written costs $2.56 in data-in charges, before queries, storage and data out.

Reviewer’s Notes: I’d start with Core if your dashboards look back less than about 72 hours, and use the At-Home license to try compaction on a hobby project. Serverless makes sense when your write volume is small and you’d rather not run a server. For commercial workloads that need months of history, get an Enterprise quote before you commit.

My Experience With InfluxDB

Installing InfluxDB 3 Core 3.12.1

I downloaded the official ~100 MB tarball for Apple Silicon (version 3.12.1, released October 15, 2026). It holds one influxdb3 binary, a bundled Python runtime for the Processing Engine, and dual MIT/Apache 2.0 license files. Core also installs via a script, Docker (influxdb:3-core) and DEB/RPM packages, and Homebrew carries 3.12.1 too.

My first start aborted: the binary auto-loads a .env file from the current directory and stopped on a line it couldn’t parse. From a clean directory, influxdb3 serve --node-id node1 --object-store file came up in 9 ms on port 8181. Object store options are memory, file, S3, Google and Azure.

The Admin Token You See Exactly Once

Auth is on by default: even /health and /ping returned 401 without a token, which I like. influxdb3 create token --admin then prints the admin token once (“will not be shown again”).

I didn’t capture mine. Running the command again failed with 409 “token name already exists”, and --regenerate itself needs the current admin token. I had to wipe the data directory and start over, so pipe that output straight into a secrets store, or use the offline option (--offline --output-file).

Core binds to 0.0.0.0:8181 by default, and until the first admin token exists, anyone who can reach the port can claim it. Bind to 127.0.0.1:8181 until the token is created.

Writing 1 Million Points

Line protocol posted to /api/v3/write_lp auto-created my database and table. My single-threaded Python client sent 1,000,000 points (100 hosts x 3 fields, 60 days of backfilled timestamps) in 10,000-line batches:

• Default durable mode: 99.9 s, about 10,000 points/s, because each batch waited ~1 s for the WAL flush
• With no_sync=true: 2.3 s, about 427,000 points/s

The catch: no_sync acknowledges before the WAL is on disk, so I’d only use it where the application tolerates losing writes that were acknowledged but not yet flushed.

High cardinality didn’t slow it down. 1,000,000 points across 200,000 unique host series took 2.3 s with no_sync, and count(DISTINCT host) returned 200,000 in 0.32 s. Writing a string into the float field usage returned HTTP 400: “invalid column type for column ‘usage’, expected float, got string”.

Hitting the 432-File Wall

After a restart persisted the data, my 1M-row table was 8,635 Parquet files (43.9 MB). Core writes one file per 10-minute block per table and has no compaction.

Querying all 60 days, or anything wider than about 2.5 to 3 days of this data, failed with “Query would scan 432 Parquet files, exceeding the file limit”, followed by advice to narrow the range, raise --query-file-limit, or upgrade to Enterprise. Inside the window, counting 44,912 rows took 0.24 s.

Backfilling 60 days is the worst case for file count, but live data hits the same wall at about 72 hours. Sparse data stretches further because the cap counts files, not hours. Raising the limit works, but InfluxData warns it slows queries and risks out-of-memory kills.

Reviewer’s Notes: Core started in 9 ms, but the one-time token cost me a full reset and persistence changed which queries succeeded. Plan for both on day one, before you load data you care about.

Querying InfluxDB 3 With SQL and InfluxQL

🔎 SQL on Apache DataFusion

SQL runs on Apache DataFusion, with date_bin for time buckets. While my 1M rows still sat in the WAL and in-memory buffer, a region-filtered daily date_bin aggregation took 1.4 s, a single-host last-3-rows lookup 0.45 s, and a full 60-day count about 8 s.

After persistence, that same count failed on the file cap, while windowed queries ran in about 0.2 s (an hourly average took 0.19 s). Test your dashboard queries after persistence, not straight after ingestion.

Both influxdb3 query and /api/v3/query_sql return results as:

• Pretty table: for reading in the terminal
• JSON or CSV: for apps and spreadsheets
• Parquet: for columnar pipelines

TimescaleDB gives you full PostgreSQL, joins and extensions included. InfluxDB’s SQL is analytical DataFusion SQL, not a PostgreSQL replacement, so I’d pick TimescaleDB when telemetry must join business tables.

📜 InfluxQL for v1 Habits

InfluxQL ran through the same influxdb3 query command and endpoint as SQL, so v1-era queries and habits still have a home in v3. Core also accepts the v1 /write and v2 /api/v2/write compatibility endpoints next to the native v3 API, so existing write clients can point at it. Write compatibility does not carry Flux queries over, though.

🚫 No Flux, No DELETE

Flux is not supported on any InfluxDB 3 product. It stays in maintenance mode on 1.x and 2.x (security patches and critical fixes only), and Flux tasks must be rewritten in SQL, InfluxQL or Python plugins. Explorer has a beta Flux-to-SQL converter, but I didn’t run it, so budget time to check every rewritten query.

SQL DELETE and UPDATE returned “DML not supported”. You remove data by dropping tables or databases (influxdb3 delete table) or with retention periods such as --retention-period 30d at database creation; the _internal database keeps 7 days by default.

Docker’s influxdb:latest tag has pointed at Core 3 since September 15, 2026, so pin influxdb:2 if you’re staying on v2.

Processing Engine and Caches: Python Inside the Database

Core ships an embedded Python runtime (Python 3.13) inside the server, so alerting, downsampling and light ETL don’t need a separate service. Plugins fire on three trigger types:

• WAL triggers: run on each WAL flush, every second by default
• Scheduled triggers: run on every: intervals or cron: expressions
• HTTP triggers: expose a plugin at /api/v3/engine/<path>

The engine needs --plugin-dir at server start, so I restarted my tarball install with it. My 10-line process_writes plugin copied any cpu row with usage > 90 into a hot_hosts table. A write with usage 97.5 landed in hot_hosts within the same second, and the log line showed up in both the server log and system.processing_engine_logs.

InfluxData’s official plugin library covers the common jobs (I didn’t run these):

• Downsampler and Gapfill: aggregate old readings and fill gaps
• Kafka, MQTT, AMQP and OPC UA subscribers: pull in streaming and industrial data
• State-change and threshold deadman checks: flag transitions and silent sensors
• Forecasting and anomaly detection: Prophet, MAD and similar models on stored series

The last value cache (default TTL 4 hours) stayed empty for my backfilled data because it only fills from new writes. After two fresh writes, last_cache('cpu','cpu_last') returned both hosts instantly.

The distinct value cache (max cardinality 100,000, max age 1 day by default) also fills from new writes. SELECT count(*) on it errored (“must either specify a row count or at least one column”), while selecting columns worked, so test cache queries before wiring them into dashboards.

TimescaleDB covers the same jobs with continuous aggregates and scheduled SQL jobs, and QuestDB stays SQL-only. Python inside the database is the feature that most sets InfluxDB 3 apart, and it’s in the free build.

Core vs Enterprise vs Cloud: Which InfluxDB 3 Do You Need?

InfluxData positions Core for recent-data, edge and single-node use, and puts HA, multi-node clusters and long-range compaction in Enterprise. Enterprise also adds commercial support and ISO 27001 and SOC 2 coverage. If your queries reach back further than about 72 hours, the real choice is between Enterprise and Cloud.

Edition	Deployment	Compaction / long-range queries	HA	Processing Engine	Price
Core	Self-managed, single node	No; 432-file query cap	No	Yes	Free
Enterprise	Self-managed, single or multi-node	Yes	Yes, with read replicas and automatic failover	Yes	Custom; At-Home free
Cloud Serverless	Managed, multi-tenant	Managed by InfluxData	Managed by InfluxData	No	Usage-based
Cloud Dedicated	Managed, single-tenant	Managed by InfluxData	Managed by InfluxData	Check with sales	Custom
Clustered	Self-managed Kubernetes	Check with sales	Check with sales	Check with sales	Custom

Cloud Serverless is not hosted Core. It uses v1/v2-compatible write endpoints with no native v3 write API and no Processing Engine, so a Core app built on Python triggers needs that logic rebuilt before it moves. I didn’t run Enterprise, Serverless, Dedicated or Clustered.

Enterprise Trial and At-Home licenses activate at first start: the server asks for your email and license type, and a verification link finishes the job. If a license expires, queries return errors and the server won’t restart, although writes and compaction continue.

Two free companion tools work with Core and Enterprise:

• Explorer 1.11.0: a web UI for Core and Enterprise with a SQL editor, charts, OIDC single sign-on, role-based access and a plugin library
• MCP server: npx -y @influxdata/influxdb3-mcp-server lets Claude, Claude Code or Cursor query and write with your InfluxDB token, so scope that token tightly before handing it to an assistant

Core went from 3.9 (April 2026) to 3.12.1 (October 2026). The 3.10 and 3.12 upgrades are one-way catalog migrations unless you keep a catalog backup, and 3.12.1 fixed a bug that could lose acknowledged writes after a forced snapshot and restart. Version 3.12 also gave --max-concurrent-queries a finite default and enforces WAL buffer limits, so writes return 429 when the buffer is full. Back up the catalog before upgrading and pin an exact version tag, because influxdb:3-core tracks the latest 3.x release.

How Does InfluxDB Compare to Competitors?

• InfluxDB vs Tiger Data (TimescaleDB): TimescaleDB is full PostgreSQL, with joins, extensions and long history on one node. Tiger Cloud starts at $30/month (Scale from $36/month) with a 30-day trial and no card, though HA replicas and other add-ons cost extra. I’d choose it when sensor readings must join customer, asset or transaction tables.

• InfluxDB vs QuestDB: QuestDB is an Apache 2.0 SQL time-series database with no equivalent of Core’s 432-file cap in its open-source build; Enterprise is contact-sales, with a free trial. I’d shortlist it for market data and long-range SQL when the database must stay open source.

• InfluxDB vs ClickHouse: ClickHouse is a columnar analytics database for any table shape, not just time series. ClickHouse Cloud bills by usage with no monthly minimum and gives $300 in trial credits. I’d favor it when telemetry is one of many analytical workloads.

• InfluxDB vs Grafana Cloud (Prometheus/Mimir): the free tier includes 10,000 active series, and Pro is a $19/month platform fee with 10,000 series included, then $6.50 per 1,000 series. I’d pick it for Kubernetes and infrastructure metrics already on Prometheus.

• InfluxDB vs Amazon Timestream for InfluxDB: AWS runs the same Core or Enterprise engine, billed per node-hour. It suits AWS-centered teams, but managed Core still has no compaction, so the file cap comes along.

InfluxDB remains my pick for ingest-heavy metrics and IoT, where Telegraf’s plugin ecosystem and Python inside the database fit the workload.

How I Reviewed InfluxDB

I installed InfluxDB 3 Core 3.12.1 from the official tarball on a 10-core Apple Silicon MacBook and wrote 3 million points in three load runs: durable writes, no_sync writes, and a 200,000-series cardinality test. I queried 60 days of data with SQL and InfluxQL before and after persistence, set up both caches, and ran a Python write-trigger plugin. These are single-client, local measurements, not a capacity forecast.

I did not run Enterprise, Cloud Serverless, Cloud Dedicated, Clustered, Explorer or the MCP server, so I describe those by what the platform does. I weighed pricing transparency, ingestion and query performance, setup and security defaults, feature depth, and fit against TimescaleDB, QuestDB and ClickHouse.

Prices current as of October 2026.

Should You Use InfluxDB for Time-Series Data?

I recommend InfluxDB for fast telemetry ingestion and Python processing inside the database, but Core is a recent-data database, not a home for months of analytics. My test paired 427,000 points/s with a 60-day query that failed after persistence.

• Yes, on Core: metrics, IoT and observability dashboards that look back hours or a few days, especially with Telegraf and Grafana in place. Keep durable writes on unless you can tolerate early acknowledgments.
• Yes, on Enterprise: months of history and HA, if you’ll take a quote.
• No: time series that must join relational data (TimescaleDB), long-range SQL in pure open source (QuestDB), or a v2 shop with heavy Flux tasks and no time to rewrite them.

Download Core, write a day of your real data plus some history, and run your longest dashboard query after persistence. If the 432-file cap blocks it, try the 30-day Enterprise trial or, for hobby use, the At-Home license.

InfluxDB FAQ

Is InfluxDB free?

Yes. InfluxDB 3 Core is free and open source under MIT/Apache 2.0. Enterprise has a free, never-expiring At-Home license capped at 2 CPU cores per cluster for non-commercial use, plus a 30-day trial. Cloud Serverless has a Free plan with 5 MB of writes per 5 minutes and 30-day retention.

What is the 72-hour limit in InfluxDB 3 Core?

It is a cap of 432 Parquet files per query. Core writes one file per 10-minute block per table and never compacts, so 432 files cover about 72 hours. It is not a retention limit: my 60-day query failed after persistence. Raise --query-file-limit carefully, or move to Enterprise for compaction.

Does InfluxDB 3 support Flux?

No. No InfluxDB 3 edition runs Flux. Flux stays in maintenance mode on InfluxDB 1.x and 2.x, and v3 uses SQL and InfluxQL. Flux tasks must be rewritten in SQL, InfluxQL or Python plugins; Explorer’s beta Flux-to-SQL converter can help, but I didn’t test it.

How much does InfluxDB Cloud cost?

Cloud Serverless charges $0.0025 per MB written, $0.012 per 100 queries, $0.002 per GB-hour stored and $0.09 per GB out, after a Free plan and a $250 starting credit. Cloud Dedicated is custom-priced by CPU, RAM and storage. Estimate all four Serverless meters, not just writes.

Is InfluxDB better than TimescaleDB?

For ingest-heavy metrics with recent-data queries and Python processing, yes. TimescaleDB is better when time series must join relational business data or need PostgreSQL extensions and long history on one node. Core’s file cap is the big weakness; Enterprise fixes it with compaction, but only at quote-only prices.