Last updated 26 June 2020

Heroku Postgres offers a wide spectrum of plans to serve use cases that range in size from personal blogs to large-dataset, high-transaction applications. Choosing the right plan depends on your app’s specific usage characteristics, including requirements for availability and uptime.

Plan tiers

Heroku Postgres’s many plans are divided into five high-level tiers. The primary difference between each tier is the amount of tolerated monthly downtime for databases in the tier. The five tiers are:

• Hobby Tier — For apps that can tolerate up to 4 hours of downtime per month
• Standard Tier — For apps that can tolerate up to 1 hour of downtime per month
• Premium Tier — For apps that can tolerate up to 15 minutes of downtime per month
• Private Tier — For Heroku Enterprise customers
• Shield Tier — For Heroku Enterprise customers who need compliance-capable databases

Here is a breakdown of the differences between the tiers:

Downtime tolerance is based on a 30-day month.

Shared features

All Heroku Postgres plan tiers share the following features:

• Fully managed database service with automatic health checks
• Write-ahead log (WAL) off-premise storage every 60 seconds, ensuring minimal data loss in case of catastrophic failure
• Daily logical database backups with PG Backups (optional but free)
• Dataclips for easy and secure sharing of data and queries
• SSL-protected psql/libpq access
• Running unmodified Postgres 9.4, 9.5, 9.6, 10, 11
• Postgres extensions
• A full-featured web UI

Hobby tier

The hobby tier, which includes the hobby-dev and hobby-basic plans, has the following limitations:

• Enforced row limits of 10,000 rows for hobby-dev and 10,000,000 for hobby-basic plans
• Maximum of 20 connections
• No in-memory cache: The lack of an in-memory cache limits performance, because the data can’t be accessed on low-latency storage.
• No fork/follow support: Fork and follow, used to create replica databases and leader-follower setups, are not supported.
• No Expensive Queries support.
• Expected uptime of 99.5% each month.
• Unannounced maintenance and database upgrades.
• No Postgres logs.

Row limit enforcement

When you exceed hobby-tier row limits and try to insert, you receive a Postgres error:

`permission denied for relation <table name>`

The row limits of hobby-tier database plans are enforced with the following mechanism:

1. When a hobby-dev database reaches 7,000 rows, or a hobby-basic database hits 7 million rows , the owner receives a warning email stating they are nearing their row limits.
2. When the database exceeds its row capacity, the owner receives an additional notification. At this point, the database receives a 7-day grace period to either reduce the number of records, or migrate to another plan.
3. If the number of rows still exceeds the plan capacity after 7 days, INSERT privileges are revoked on the database. Data can still be read, updated, or deleted from the database. This ensures that users can bring their database into compliance and retain access to their data.
4. When the number of rows is again in compliance with the plan limit, INSERT privileges are automatically restored to the database. Note that database sizes are checked asynchronously, so it may take a few minutes for privileges to be restored.

Standard tier

The Standard tier is designed for production applications that can tolerate up to 1 hour of downtime in any given month. All standard tier databases include:

• No row limitations
• Increasing amounts of in-memory cache
• Fork and follow support
• Rollback up to 4 days
• Database metrics published to application log stream for further analysis
• Priority service restoration on disruptions
• Automatic encryption-at-rest of all data written to disk

Within the Standard tier, plans have differing memory, connection limits, and storage limits. The plans for the standard tier are:

Premium tier

The Premium tier is designed for production applications that can tolerate up to 15 minutes of downtime in any given month. All premium tier databases include:

• No row limitations
• Increasing amounts of in-memory cache
• Fork and follow support
• Rollback up to 7 days
• Database metrics published to application log stream for further analysis
• Priority service restoration on disruptions
• Automatic encryption-at-rest of all data written to disk.

Within the Premium tier, plans have differing memory, connection limits, and storage limits. The plans for the premium tier are:

Private and Shield Tiers

Heroku offers Heroku Postgres in private spaces for Heroku Enterprise customers. Additionally, Postgres shield plans are available for customers who need compliance-capable databases. For details on our private and shield plans, see the Heroku Postgres and Private Spaces article.

Cache size

Each production tier plan’s RAM size indicates the total amount of System Memory on the underlying instance’s hardware, most of which is given to Postgres and used for caching. Although a small amount of RAM is used for managing connections and other tasks, Postgres takes advantage of almost all of this RAM for its cache. Learn more about how this works in this article.

Postgres constantly manages the cache of your data: rows you’ve written, indexes you’ve made, and metadata Postgres keeps. When the data required for a query is contained entirely in the cache, performance is very fast. Queries made from cached data are often 100-1000x faster than those made from the full data set.

Conversely, having to fall back to disk is at least an order of magnitude slower. Additionally, columns with large data types (e.g. large text columns) are stored out-of-line via TOAST, and accessing large amounts of TOASTed data can be slow.

General guidelines

Access patterns vary greatly from application to application. Many applications access only a small, recently modified portion of their overall data. Postgres can always keep that portion in its cache as time goes on, and consequently these applications can perform well on smaller plans.

Applications that frequently access all of their data don’t have this luxury. These apps can observe dramatic increases in performance by ensuring that their entire dataset fits in memory. To determine the total size of your dataset, use the heroku pg:info command and look for the Data Size row:

$ heroku pg:info
=== HEROKU_POSTGRESQL_CHARCOAL_URL (DATABASE_URL)
Plan:        Crane
Status:      available
Data Size:   9.4 MB
...

Although a crude measure, choosing a plan that has at least as much in-memory cache available as the size of your total dataset ensures high cache ratios. However, you will eventually reach the point where you have more data than the largest plan, and you will have to shard. Plan ahead for sharding: it takes a long time to execute a sharding strategy.

Determining required cache size

There is no substitute for observing the database demands of your application with live traffic to determine the appropriate cache size. Ideally, your cache hit ratio should be in the 99%+ range. Uncommon queries should require less than 100ms and common ones less than 10ms.

To measure the cache hit ratio for tables:

SELECT
    'cache hit rate' AS name,
     sum(heap_blks_hit) / (sum(heap_blks_hit) + sum(heap_blks_read)) AS ratio
FROM pg_statio_user_tables;

or the cache hit ratio for indexes:

SELECT
    'index hit rate' AS name,
    (sum(idx_blks_hit)) / sum(idx_blks_hit + idx_blks_read) AS ratio
FROM pg_statio_user_indexes

Both queries should indicate a ratio near 0.99:

heap_read | heap_hit |         ratio
-----------+----------+------------------------
       171 |   503551 | 0.99966041175571094090

When the cache hit ratio begins to decrease, upgrading your database generally brings the ratio back up to 99%. To upgrade, it is recommended to perform a follower changeover update to move between plans. Watch New Relic, and see what works best for your application’s access patterns.