Software Heritage - Object storage

Content-addressable object storage for the Software Heritage project.

Quick start

The easiest way to try the swh-objstorage object storage is to install it in a virtualenv. Here, we will be using virtualenvwrapper but any virtual env tool should work the same.

In the example below we will create a new objstorage using the pathslicer backend.

~/swh$ mkvirtualenv -p /usr/bin/python3 swh-objstorage
[...]
(swh-objstorage) ~/swh$ pip install swh.objstorage
[...]
(swh-objstorage) ~/swh$ cat >local.yml <<EOF
objstorage:
  cls: pathslicing
  root: /tmp/objstorage
  slicing: 0:2/2:4/4:6
EOF
(swh-objstorage) ~/swh$ mkdir /tmp/objstorage
(swh-objstorage) ~/swh$ swh objstorage -C local.yml rpc-serve -p 15003
INFO:swh.core.config:Loading config file local.yml
======== Running on http://0.0.0.0:15003 ========
(Press CTRL+C to quit)

Now we have an API listening on http://0.0.0.0:15003 we can use to store and retrieve objects from. In an other terminal, you can import all the files from a local directory in this objstorage:

~/swh$ workon swh-objstorage
(swh-objstorage) ~/swh$ cat >remote.yml <<EOF
objstorage:
  cls: remote
  url: http://127.0.0.1:15003
EOF
(swh-objstorage) ~/swh$ swh objstorage -C remote.yml import .
INFO:swh.core.config:Loading config file remote.yml
Imported 1369 files for a volume of 722837 bytes in 2 seconds

Winery developer’s check-list

Working on Winery, the production backend, requires a slightly longer set-up.

First ensure your virtualenv contains the correct dependencies:

pip install -e .[winery]

Then create a postgres DB, called winery:

swh db create -d winery objstorage.backends.winery
swh db init -d winery objstorage.backends.winery

Prepare a container folder:

mkdir /home/martin/objstores/winery

And set it in a configuration file we’ll call localwinery.yml:

objstorage:
  cls: winery

  # boolean (false (default): allow writes, true: only allow reads)
  readonly: false
  rw_idle_timeout: 300

  database:
    # string: PostgreSQL connection string for the object index and read-write shards
    db: "dbname=winery"

    # string: PostgreSQL application name for connections (unset by default)
    application_name: localwinery

  shards_pools:
    - ## Settings for a directory pool storing swh-shards
      # Shards are stored in `{base_directory}/{pool_name}`
      type: directory
      base_directory: /home/martin/objstores/winery
      pool_name: the-shards
      # integer: threshold in bytes above which shards get packed. Can be
      # overflowed by the max allowed object size.
      shard_max_size: 100_000_000  # 100MB
  shard_active_pool: the-shards

  packer:
    # Whether the packer should create shards in the shard pool, or defer to
    # the pool manager (default: true, the packer creates images)
    create_images: true

Note that you have to run a packer and a cleaner process separately. You might want to use a smaller max_size to trigger the packer more frequently.

Now you’ll need a few terminal splits/tabs because we’ll start 3 services

# Main service (winery writer)  listens on 0.0.0.0:15003
swh objstorage -C localwinery.yml rpc-serve -p  15003
# Winery Packer Service
swh objstorage -C localwinery.yml winery packer
# optional, relevant later: RW Shard Cleaner
swh objstorage -C localwinery.yml winery rw-shard-cleaner

To import contents we’ll use the swh objstorage import, with the remote.yml configuration created in the Quick Start section in order to use the RPC server we’ve just started:

swh objstorage -C remote.yml import ~/swh-environment/

Test dependencies

Some tests do require non-python dependencies to be installed on the machine.

Azurite

The azurite tool is needed for Azure backend tests. Since it’s a npm package, you can install it using:

~/swh$ npm install azurite

and run tests with:

~/swh$ AZURITE_EXE=$HOME/node_modules/azurite/dist/src/blob/main.js tox

Ceph

Some Winery tests suites really manipulate RBD images, but this requires the ceph binary, configured to access a realistic cluster, and an explicit environment variable to flag you really want to run this suite (it takes a few minutes). Otherwise these tests are skipped.

To run these tests, on a developer’s machine you can install MicroCeph with snap. But it has permissions issues so you also need binaries from the Debian package and a hacky configuration. So we wrapped it in a Bash script. You need to pre-install the snap and ceph-common Debian packages (or your distribution’s equivalent), then run:

(swh) ~/swh-environment/swh-objstorage$ ./bin/test_winery_with_microceph.sh

This script uses the $PYTEST_FLAGS environment variable, that you can set for example to -v or -x.

Metadata

Release files for swh.objstorage 7.4.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for swh.objstorage 7.4.1
File	Size	Uploaded
swh_objstorage-7.4.1.tar.gz	140.5 kB	Oct 16, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for swh.objstorage 7.4.1
File	Interpreter	ABI	Platform	Reset
swh_objstorage-7.4.1-py3-none-any.whl 157.6 kB Oct 16, 2026	Python 3	none	any	Details

Total release size: 298.1 kB

Release files / swh_objstorage-7.4.1.tar.gz

Download URL	swh_objstorage-7.4.1.tar.gz
Size	140.5 kB
Tags	Source
SHA-256 checksum How to use checksums	`450da5f225ba5f6287149320e3431699c14bfe2e218558924b73b1be28f50e66`
BLAKE2b-256 checksum How to use checksums	`c1d8f85f1d68da2995780a4a7108fc024cbc86cf6b89d6a06a3fc62fa97039b3`
Upload date	Oct 16, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.11.15`

Release files / swh_objstorage-7.4.1-py3-none-any.whl

Download URL	swh_objstorage-7.4.1-py3-none-any.whl
Size	157.6 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`fbce249fecb9e39e648aa6a586c0c65c563aa42f60e6e9b12fdeee0f8622630a`
BLAKE2b-256 checksum How to use checksums	`4ba0f7cb3249b09fc3bc246a05a951163494b18ca409eda909d96a02ffec1123`
Upload date	Oct 16, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.11.15`