Persistence

Related pages

• Configuration: Backups

Overview

When running Weaviate with Docker or Kubernetes, you can persist its data by mounting a volume to store the data outside of the containers. Doing so will cause the Weaviate instance to also load the data from the mounted volume when it is restarted.

Note that Weaviate now offers native backup modules starting with v1.15 for single-node instances, and v1.16 for multi-node instances. For older versions of Weaviate, persisting data as described here will allow you to back up Weaviate.

Docker Compose

Persistence

When running Weaviate with Docker Compose, you can set the volumes variable under the weaviate service and a unique cluster hostname as an environment variable.

services:
  weaviate:
    volumes:
      - /var/weaviate:/var/lib/weaviate
    environment:
      CLUSTER_HOSTNAME: 'node1'

• About the volumes
  • /var/weaviate is the location where you want to store the data on the local machine
  • /var/lib/weaviate (after the colon) is the location inside the container, don't change this
• About the hostname
  • The CLUSTER_HOSTNAME can be any arbitrarily chosen name

In the case you want a more verbose output, you can change the environment variable for the LOG_LEVEL

services:
  weaviate:
    environment:
      LOG_LEVEL: 'debug'

A complete example of a Weaviate without modules but with an externally mounted volume and more verbose output:

---
version: '3.4'
services:
  weaviate:
    command:
    - --host
    - 0.0.0.0
    - --port
    - '8080'
    - --scheme
    - http
    image: semitechnologies/weaviate:1.22.5
    ports:
    - 8080:8080
    restart: on-failure:0
    volumes:
      - /var/weaviate:/var/lib/weaviate # <== set a volume here
    environment:
      QUERY_DEFAULTS_LIMIT: 25
      AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED: 'true'
      PERSISTENCE_DATA_PATH: '/var/lib/weaviate'
      DEFAULT_VECTORIZER_MODULE: 'none'
      ENABLE_MODULES: ''
      CLUSTER_HOSTNAME: 'node1' # <== this can be set to an arbitrary name
...

Backups

See Backups.

Kubernetes

For Kubernetes setup, the only thing to bear in mind is that Weaviate needs a PersistentVolumes through PersistentVolumeClaims (more info) but the Helm chart is already configured to store the data on an external volume.

Disk Pressure Warnings and Limits

Starting with v1.12.0 there are two levels of disk usage notifications and actions configured through environment variables. Both variables are optional. If not set they default to the values outlined below:

Variable	Default Value	Description
DISK_USE_WARNING_PERCENTAGE	80	If disk usage is higher than the given percentage a warning will be logged by all shards on the affected node's disk
DISK_USE_READONLY_PERCENTAGE	90	If disk usage is higher than the given percentage all shards on the affected node will be marked as READONLY, meaning all future write requests will fail.

If a shard was marked READONLY due to disk pressure and you want to mark the shard as ready again (either because you have made more space available or changed the thresholds) you can use the Shards API to do so.

Virtual memory access method

Added in v1.21

You can choose between mmap (DEFAULT) and pread functions to access virtual memory by setting the PERSISTENCE_LSM_ACCESS_STRATEGY environment variable.

The two functions reflect different under-the-hood memory management behaviors. mmap uses a memory-mapped file, which means that the file is mapped into the virtual memory of the process. pread is a function that reads data from a file descriptor at a given offset.

In general, mmap may be a preferred option with memory management benefits. However, if you experience stalling situations under heavy memory load, we suggest trying pread instead.