TimescaleDB vs InfluxDB
TimescaleDB vs InfluxDB compared for 2026: TimescaleDB 2.32 as a PostgreSQL extension with hypertables, the hypercore columnstore, continuous aggregates and the Apache vs Community (TSL) editions vs InfluxDB 3 Core and Enterprise with object-storage Parquet, SQL and InfluxQL, the Python processing engine, Core's 72-hour query window and database limits, licences, Tiger Cloud and InfluxDB Cloud pricing, and a clear pick per situation.
Quick Answer
TimescaleDB and InfluxDB both store time-series data, but they are different kinds of system. TimescaleDB is a PostgreSQL extension: you keep full SQL, joins with relational tables, transactions and the Postgres ecosystem, and add hypertables that partition by time, a hybrid row-columnar engine (hypercore) with 90 to 98% compression by Tiger Data's figures, continuous aggregates and retention policies. Version 2.32.0 shipped on 12 November 2026 for PostgreSQL 16, 17 and 18. The catch is the licence split: the free, self-hostable Community Edition (Timescale License) has those features, while the Apache-2.0 edition lacks the columnstore, continuous aggregates and retention policies, and nobody but Tiger Data may sell Community as a service. InfluxDB 3 is a purpose-built engine that writes Parquet to object storage, speaks line protocol, SQL and InfluxQL, and runs Python plugins inside the database. Its open-source Core (3.12.2, 2 November 2026) is aimed at recent data: by default a query can touch 432 Parquet files, about 72 hours, and Core caps you at 5 databases and 2,000 tables. Historical queries, HA and read replicas need Enterprise, which is priced by CPU on request (a free At-Home licence covers 2 CPUs, non-commercial). Pick TimescaleDB for SQL analytics over long history; pick InfluxDB for metrics and IoT pipelines built around line protocol and Telegraf.
Last updated November 15, 2026. Covers TimescaleDB 2.32.0 (12 November 2026) and 2.31.0, and InfluxDB 3 Core and Enterprise 3.12.2 (2 November 2026) with 3.11.8, using Tiger Data docs (editions matrix, hypertables, hypercore, tiered storage, limitations, supported platforms, install), the Tiger Cloud pricing page, the timescaledb LICENSE and TSL, GitHub release notes, docs.influxdata.com (Core and Enterprise overviews, configuration options, database limits, retention, processing engine, SQL and InfluxQL references, licence types, v1/v2 migration, release notes), the InfluxDB pricing page and the GitHub API, all read on 15 November 2026.
| Feature | TimescaleDB | InfluxDB |
|---|---|---|
| Architecture | PostgreSQL extension (hypertables partitioned by time into chunks) | Purpose-built time-series engine (InfluxDB 3), diskless with Parquet persistence |
| License | Apache-2.0 outside the tsl directory; Timescale License (TSL) for Community features | Core: MIT or Apache-2.0; Enterprise: commercial licence |
| Free Self-Hosted Edition | Community Edition (TSL): free to self-host and modify, not to resell as a service; Apache 2 Edition with a reduced feature set | InfluxDB 3 Core (recent-data focus); Enterprise At-Home licence: 2 CPUs, single node, non-commercial |
| Query Languages | Full PostgreSQL SQL plus hyperfunctions (time_bucket, gapfill, percentile and time-weight functions) | SQL (Apache Arrow DataFusion) and InfluxQL (partial on v3) |
| Ingest Protocols | SQL INSERT and COPY via any PostgreSQL client | Line protocol, including v2 (/api/v2/write) and v1 (/write) compatible endpoints; Telegraf |
| Storage Engine | Hypercore: rowstore for new data, automatic conversion to columnstore | WAL plus Parquet files in object storage or on local disk |
| Compression | Columnstore compression, 90-98% per Tiger Data; Adaptive Integer Compression on by default since 2.32.0 | Columnar Parquet files; no ratio quoted here |
| Querying Long History | Yes, no documented query window | Core: about 72 hours by default (query-file-limit 432); Enterprise: historical queries supported |
| Downsampling & Rollups | Continuous aggregates with refresh policies (Community) | Python processing-engine plugins (scheduled triggers) rather than a built-in continuous-aggregate feature |
| Retention | add_retention_policy and drop_chunks (policy is Community only) | Per-database retention period (default none, minimum 1 hour), enforced at query time |
| Updates & Deletes | INSERT, UPDATE, DELETE and UPSERT, including on the columnstore (not across chunks) | Overwrite by writing the same series and timestamp; row-level delete listed as coming soon for Enterprise |
| Joins with Relational Data | Yes, standard SQL joins with any PostgreSQL table | SQL joins within InfluxDB only |
| In-Database Processing | Job scheduler (add_job) and PostgreSQL functions; Tiger Cloud also lists vector search and AI workflows | Embedded Python VM: triggers on write (WAL), on a schedule or on HTTP request (Python 3.13.18 in 3.12.2) |
| Documented Limits | Hypertable rules (non-NULL partition column, unique indexes include it, no hypertable-to-hypertable FKs); no table or database caps | Core: 5 databases, 2,000 tables total, 500 columns per table |
| High Availability & Replicas | PostgreSQL streaming replication self-hosted; Tiger Cloud HA replicas (1 on Performance, up to 2 on Scale) and read replicas on Scale | Enterprise only: HA, read replicas, dedicated ingest, query, compactor and process nodes |
| Object Storage / Tiering | Tiger Cloud tiered storage to S3/Azure Blob in Parquet, $0.021/GB-month on Scale and Enterprise | Native: S3, Google Cloud Storage, Azure Blob, local file or memory |
| Platforms | Linux, Windows, macOS; Docker and Kubernetes; PostgreSQL 16-18 for 2.32 | Linux, macOS, Windows; Docker, DEB and RPM |
| Managed Cloud | Tiger Cloud on AWS and Azure (Performance, Scale, Enterprise) | InfluxDB Cloud Serverless, Cloud Dedicated, Amazon Timestream for InfluxDB |
| Cloud Pricing (list) | Performance: compute from $30/month, storage $0.177/GB-month; Scale: from $36/month, $0.212/GB-month (effective prices assuming 5x compression, per Tiger Data) | Serverless: $0.0025/MB written, $0.012 per 100 queries, $0.002/GB-hour stored, $0.09/GB out; Dedicated custom |
| Self-Managed Enterprise | TimescaleDB Enterprise, self-managed, in public beta | InfluxDB 3 Enterprise, licensed by CPU (batches of 8 to 128 cores), custom price |
| Compatibility & Migration | Anything that speaks PostgreSQL | InfluxDB v1 and v2 write APIs, v1 /query API for InfluxQL, v1/v2 client libraries |
| Latest Release (Nov 2026) | 2.32.0 (12 Nov 2026) | 3.12.2 (2 Nov 2026); 3.11.8 maintenance; 1.13.2 (30 Oct 2026) |
TimescaleDB
TimescaleDB is an extension that turns PostgreSQL into a time-series and real-time analytics database. Hypertables partition tables by time into chunks automatically, the hypercore engine moves cooling data from rowstore to a compressed columnstore, and continuous aggregates keep rollups up to date incrementally. It is developed by Tiger Data, the company behind the timescale/timescaledb project, which also runs Tiger Cloud on AWS and Azure and has a self-managed TimescaleDB Enterprise in public beta.
Pros
- It is PostgreSQL: full SQL, joins with your relational tables, transactions, and every Postgres driver, ORM and tool works
- Hypercore columnstore compresses cold data (Tiger Data cites 90 to 98%) and still allows INSERT, UPDATE, DELETE and UPSERT on compressed chunks
- Continuous aggregates, retention policies and a job scheduler are built in (Community Edition)
- No query window or table-count caps in the free self-hosted edition; long-history analytics is the core use case
- Community Edition is free to self-host in production, including modifying the source
- Tiger Cloud publishes prices: compute from $30 a month, 30-day free trial
Cons
- Columnstore, continuous aggregates, retention policies, jobs and most hyperfunctions are Community (Timescale License) only, not in the Apache-2.0 build
- You cannot offer TimescaleDB Community Edition as a hosted database service; many third-party Postgres hosts ship only the Apache edition
- Hypertable limits: partitioning columns cannot be NULL, unique indexes must include them, and updates or upserts that move rows between chunks are not supported
- Ingest uses SQL (INSERT or COPY); there is no native line protocol endpoint, so line-protocol pipelines need a PostgreSQL-compatible writer
- You operate PostgreSQL too: vacuum, connection pooling and major-version upgrades are your job when self-hosting
Best For
Teams that want time-series and event analytics inside PostgreSQL, with SQL joins against business data, long retention, rollups, and updates or late-arriving data, self-hosted for free or on Tiger Cloud.
InfluxDB
InfluxDB is InfluxData's purpose-built time-series database. The current generation, InfluxDB 3, is a diskless engine that persists data as Parquet files on object storage or local disk, queries with SQL (Apache DataFusion) and InfluxQL, accepts line protocol through v1 and v2 compatible write APIs, and embeds a Python processing engine for plugins and triggers. InfluxDB 3 Core is the open-source release; InfluxDB 3 Enterprise adds historical queries, high availability, read replicas and multi-node clusters. InfluxData also runs InfluxDB Cloud Serverless and Cloud Dedicated.
Pros
- Built for high-rate metrics and sensor ingest with line protocol, plus write compatibility with InfluxDB v1 and v2 clients and Telegraf
- Diskless design: Parquet on S3, Google Cloud Storage, Azure Blob or local disk, so storage scales independently of compute
- Last-value and distinct-value caches for dashboard lookups (InfluxData cites under 10 ms and 30 ms)
- Embedded Python processing engine runs plugins on writes, on schedules or on HTTP requests inside the database
- Core is open source and runs on Linux, macOS, Windows and Docker; Enterprise has a free 30-day trial and a free At-Home licence
- Cloud Serverless publishes usage prices and gives $250 of credit
Cons
- Core is designed for recent data: with defaults a query can read 432 Parquet files, roughly 72 hours, and raising the limit risks out-of-memory kills
- Core caps you at 5 databases, 2,000 tables across them and 500 columns per table
- Historical queries, HA and read replicas require Enterprise, whose price is custom and set by CPU count
- InfluxQL support on the v3 engine is still being rebuilt and some features are not implemented yet
- No relational joins against business data held in another database; it is a separate system to run next to your primary store
Best For
Metrics, IoT and observability pipelines built around line protocol and Telegraf, especially recent-data dashboards and alerting, or teams already on InfluxDB 1.x or 2.x moving to the v3 engine.
Architecture: a Postgres extension versus a purpose-built engine
TimescaleDB does not replace PostgreSQL, it extends it. A hypertable looks and behaves like a normal table, but the extension splits it into chunks by time (and optionally other dimensions), creates a descending time index automatically and prunes chunks at query time. Hypercore then stores new chunks in the rowstore for fast inserts and updates and converts them to a columnstore as they cool, with chunk skipping, vectorized aggregation, bloom and min-max sparse indexes, and summary queries answered from batch metadata. Because it is still Postgres, you get transactions, joins against any table, extensions and every driver. InfluxDB 3 is its own engine, written around Apache Arrow, DataFusion and Parquet. Writes go to a write-ahead log and are persisted as Parquet files on object storage (S3, Google Cloud Storage, Azure Blob) or local disk, which InfluxData calls a diskless architecture, and queries run over those files plus in-memory caches. The practical consequence: TimescaleDB is one database for both your relational and time-series data, while InfluxDB is a separate, specialised store that you feed from agents and query alongside your primary database. Neither is wrong; it depends on whether joining with business data matters.
Licences and what the free editions really include
Both vendors split features by licence, and the splits matter more than the headline 'open source'. TimescaleDB's repository is Apache-2.0 outside the tsl directory and Timescale License inside it. The Apache 2 Edition gives you hypertables, chunks, time_bucket and basic informational views; the Community Edition, under the TSL, adds the columnstore (hypercore), continuous aggregates, retention policies, the job scheduler, gapfill, interpolation and the percentile, time-weight and sketch hyperfunctions. Community is free to self-host and modify in production, but you may not offer it as a database service, so a managed Postgres provider may ship only the Apache build; run SHOW timescaledb.license to see which you have. InfluxDB 3 Core is open source (the repo carries MIT and Apache-2.0 licence files) with no feature licence key, but it is deliberately scoped to recent data. InfluxDB 3 Enterprise is commercial and licensed per cluster by CPU count: a 30-day Trial (up to 256 CPUs, multi-node, commercial use), a free At-Home licence (2 CPUs, never expires, but no multi-node and no commercial use) and a Commercial licence bought in batches of 8, 16, 32, 64 or 128 cores.
Long-range queries and limits: the 72-hour question
The single biggest surprise for teams evaluating InfluxDB 3 Core is its query window. Core's query-file-limit defaults to 432 Parquet files, and with the default 10-minute generation-1 duration that lets a query reach up to about 72 hours of data. You can raise the limit, but InfluxData warns of degraded query performance, higher memory use and the process being killed for running out of memory. Core also limits you to 5 databases, 2,000 tables across all databases and 500 columns per table. Enterprise lifts the history restriction (InfluxData lists historical query capability and single-series indexing as Enterprise additions) and adds compaction across dedicated nodes. TimescaleDB has no equivalent window: querying a year of data is a normal SQL query over compressed chunks, and the documented constraints are structural rather than size caps. Hypertable partitioning columns cannot contain NULLs or be generated columns, unique indexes must include every partitioning column, updates that would move a row into another chunk (including upserts with ON CONFLICT UPDATE) are not supported, and a hypertable cannot hold a foreign key to another hypertable. If dashboards over the last day are your whole workload, Core's window is fine; if analysts will ask about last quarter, plan for TimescaleDB or InfluxDB Enterprise.
Rollups, retention and in-database processing
TimescaleDB handles downsampling with continuous aggregates: materialized views over time_bucket queries that refresh incrementally on a policy, can themselves be compressed, and combine with retention policies that drop raw chunks after a period while keeping the rollups. Version 2.32.0 improved refresh performance by updating the watermark once per batch and turned on Adaptive Integer Compression by default, which Tiger Data says improves integer compression ratios by up to 20%. Background work runs through the built-in job scheduler, and 2.32 lets database owners manage background workers without superuser. InfluxDB 3 takes a programmable route. Each database can have a retention period (none by default, as short as one hour), enforced at query time with physical deletion afterwards. Downsampling, alerting and transformations are written as Python plugins for the embedded processing engine, triggered on writes, on a schedule (for example every:1m) or by HTTP requests, with an official plugins repository to start from. The last-value cache and distinct-value cache serve the latest reading per series or distinct tag values for dashboards. TimescaleDB's approach is declarative SQL; InfluxDB's is code next to the data. Teams with Python skills and event-driven needs may prefer the latter.
Managed services and what they cost
Tiger Cloud, on AWS and Azure, publishes list prices. The Performance plan starts at $30 a month for compute with storage at $0.177 per GB-month, up to 4 services of 8 CPUs and 32 GB each, one HA replica and 3 days of point-in-time recovery. Scale starts at $36 a month with storage at $0.212 per GB-month, unlimited services up to 32 CPUs and 128 GB, read replicas, up to two HA replicas, 14 days of recovery and tiered storage to S3 or Azure Blob in Parquet at $0.021 per GB-month, charged on the original uncompressed size. Tiger Data describes its storage prices as effective prices assuming 5x compression, so check how a quote applies to your data. Enterprise adds SAML SSO, HIPAA and 24x7 support at a custom price, and every plan has a 30-day free trial. InfluxDB Cloud Serverless charges $0.0025 per MB written, $0.012 per 100 query executions, $0.002 per GB-hour stored and $0.09 per GB transferred out, with $250 of credit; it is multi-tenant with service quotas and no HA. As list-price arithmetic, keeping 100 GB for a 730-hour month costs about $146 in storage alone, before writes and queries. Cloud Dedicated and Enterprise are quoted per customer, and Amazon Timestream for InfluxDB bills through AWS.
Verdict
Pick TimescaleDB when your time-series data needs to live next to relational data and be queried with ordinary SQL over months or years: product analytics, financial ticks joined to instruments, IoT readings joined to device and customer tables, anything with late-arriving rows, corrections or upserts. Self-host the Community Edition for free (it has the columnstore, continuous aggregates and retention policies the Apache build lacks), or use Tiger Cloud from $30 a month for compute if you would rather not run PostgreSQL. Check your hosting first, because only Tiger Data may offer the Community Edition as a service. Pick InfluxDB when the workload is a metrics or sensor pipeline built on line protocol and Telegraf, where most queries hit the last few hours or days, dashboards need instant last-value lookups, and you like storage on cheap object storage with Python plugins next to the data. InfluxDB 3 Core is a good free fit for that recent-data window, but be honest about history: Core's default query reaches about 72 hours and it is capped at 5 databases, so long-range analysis, HA and read replicas mean InfluxDB 3 Enterprise at a custom per-CPU price, or Cloud Dedicated. If you are on InfluxDB 1.x or 2.x today and like it, the v3 engine's compatible write and InfluxQL APIs make staying in the family the easier path; if you are starting fresh and already run PostgreSQL, TimescaleDB is usually the simpler stack.