[pbs-devel] [PATCH proxmox{, -backup} v10 00/49] fix #2943: S3 storage backend for datastores
Christian Ebner
c.ebner at proxmox.com
Mon Jul 21 18:44:18 CEST 2025
Disclaimer: These patches are still in an experimental state and not
intended for production use.
This patch series aims to add S3 compatible object stores as storage
backend for PBS datastores. A PBS local cache store using the regular
datastore layout is used for faster operation, bypassing requests to
the S3 api when possible. Further, the local cache store allows to
keep frequently used chunks and is used to avoid expensive metadata
updates on the object store, e.g. by using local marker file during
garbage collection.
Backups are created by upload chunks to the corresponding S3 bucket,
while keeping the index files in the local cache store, on backup
finish, the snapshot metadata are persisted to the S3 storage backend.
Snapshot restores read chunks preferably from the local cache store,
downloading and insterting them if not present from the S3 object
store. Listing and snapsoht metadata operation currently rely soly on
the local cache store.
Currently chunks use a 1:1 mapping to S3 objects. An advanced packing
mechanism for chunks to significantly reduce the number of api
requests and therefore be more cost effective will be implemented as
followup patches.
Most notably changes since version 8 of the patches (thanks @Lukas for
more testing and debugging and Hannes for review):
- Moved s3 refresh button to the datastore content, and made it less
visible by placing it into a 'More' dropdown.
- Added basic unit tests for s3 object key helpers
- Extend missing public function and type documentation
- Use pbs_buildcfg::configdir macro to build config paths
- Mostly refactoring based on feedback, please refer to the per-patch
changelog for details.
- Fixes typos in docs (thanks @Maximiliano)
Most notably changes since version 8 of the patches (thanks @Lukas for
code review):
- Moved s3 refresh button to the datastore content, and made it less
visible by placing it into a 'More' dropdown.
- Added basic unit tests for s3 object key helpers
- Extend missing public function and type documentation
- Use pbs_buildcfg::configdir macro to build config paths
- Mostly refactoring based on feedback, please refer to the per-patch
changelog for details.
- Fixes typos in docs (thanks @Maximiliano)
Most notably changes since version 7 of the patches (thanks @Thomas
and @Lukas for feedback, testing and debugging):
- Improve self-signed certificate fingerprint check, verify valid
expected fingerprint is passed to client on instantiation.
- Rename previously missed host to endpoint is s3 client selector
- Use more specific `S3 Client ID` over ambiguous `Unique Identifier`
- Implement missing cli commands for s3 client manipulation
- Add in-use marker to s3 object stores to avoid accitental reuse of
object stores which are already used as datastore by another
instance, adding also flags to overwrite.
- Automatically perform an s3-refresh when recreating a datastore,
pre-populating the contents without further user interaction.
- Add documentation
- Fix formatting issue in proxmox-s3-client
Most notably changes since version 6 of the patches (thanks @Thomas
for feedback):
- Reworked uri encoding logic, instead of doing this in the
S3ObjectKey, perform this in the build_uri helper used by all
client api requests.
- Add cache-size optional parameter to datastore backend config,
allows to define the local datastore LRU cache capacity.
- Increase s3 client timeout, as otherwise delete objects operations
on Cloudflare R2 would run into a timeout error.
- Also upload client log, previously not uploaded to s3 backend.
- Add missing documentation to some pub types in the response reader
- Use s3 object key generation helper for index file upload, which
fixes the missing key prefix.
- Add basic regression tests for uri encoder and decoder helper
functions.
- Include some baseline performance tests for garbage collection as
well as chunk up-/download when caching.
Most notably changes since version 5 of the patches (thanks @Thomas
for feedback):
- Move s3 client into its own, dedicated crate in the proxmox repo
- Factor out any directly PBS related code from the client
- Guard implementation behind feature cfg, so api types can be used
independently
- Add basic example and extend on crate documentation
Most notably changes since version 4 of the patches:
- Fix race between S3 backend upload and local cache store insert,
avoiding possibly chunk loss for concurrent backups.
- Use the local datastore cache also for local chunk reader instances
- Fallback to fetching chunks from S3 backend if they should be cached
but the local chunk file is missing or empty, instead of failing
- Rename chunks detected as corrupt also on the S3 object store
- Retry chunk uploads via put objects in case of errors.
- Add possibility to add rate limits for the s3 client put requests, as
otherwise object stores can be overloaded.
- Allow for Cloudflare R2 compatible `auto` region, as otherwise AWS
sign v4 request authentication will fail
- Use `Async` instead of `Sync` variant for the api handler of the
s3-refresh command, as otherwise this fails.
- Take into account that some type folders might not be present when
performing an s3-refresh.
- Use `Local` instead of `Regular` to refer to normal datastores in the
creation window.
Most notably changes since version 3 of the patches:
- Rebased onto current master, fixed incompatibilities with upgraded
dependencies
- Added method to uri decode s3 object keys, as they are required in
order to download contents to a local store
- Added api endpoint to allow resyncing of the datastore contents to
the local cache store, introducing a new maintenance mode s3-refresh
to guarantee consistency.
Most notably changes since RFC version 2 of the patches (thanks
@Lukas for feedback):
- Extend S3 client implementation to also support path style bucket
addressing.
- Keep bucket name as config option for the datastore, allowing more
flexible reuse of a configured S3 client.
- Use the datastore name as additional object key prefix to allow for
multiple datastores on the same bucket.
- Allow bucket and region templating in S3 endpoint, making this more
flexible with respect to possible DNS records.
- Rework datastore create window to be less overloaded.
- Drop dead code in the S3 client implementation, since tagging and
object copying is currently not required.
- Fix missing locking when deleting chunks from s3 store during
garbage collection, avoiding possible chunk loss for concurrent
backups.
- Remove chunks from LRU cache when deleting chunks during garbage
collection, avoiding possible chunk loss for concurrent backups.
- Add dedicated types for object prefix and relative s3 key paths to
avoid misuse.
- Use more fitting icon for S3 client.
Link to the bugtracker issue:
https://bugzilla.proxmox.com/show_bug.cgi?id=2943
Steps to setup a local S3 object store using RADOS gateway or MinIO
can be found at (internal only, external users might use the steps
outlined in the cover letter and comments of RFC version 2):
https://wiki.intra.proxmox.com/PBS_Setup_S3_Object_Store
proxmox:
Christian Ebner (3):
pbs-api-types: extend datastore config by backend config enum
pbs-api-types: maintenance: add new maintenance mode S3 refresh
s3 client: Add missing S3 object key max length check
Cargo.toml | 1 +
pbs-api-types/Cargo.toml | 1 +
pbs-api-types/debian/control | 2 +
pbs-api-types/src/datastore.rs | 114 +++++++++++++++++++++++-
pbs-api-types/src/maintenance.rs | 4 +
proxmox-s3-client/examples/s3_client.rs | 4 +-
proxmox-s3-client/src/object_key.rs | 26 ++++--
7 files changed, 142 insertions(+), 10 deletions(-)
proxmox-backup:
Christian Ebner (46):
datastore: add helpers for path/digest to s3 object key conversion
config: introduce s3 object store client configuration
api: config: implement endpoints to manipulate and list s3 configs
api: datastore: check s3 backend bucket access on datastore create
api/cli: add endpoint and command to check s3 client connection
datastore: allow to get the backend for a datastore
api: backup: store datastore backend in runtime environment
api: backup: conditionally upload chunks to s3 object store backend
api: backup: conditionally upload blobs to s3 object store backend
api: backup: conditionally upload indices to s3 object store backend
api: backup: conditionally upload manifest to s3 object store backend
api: datastore: conditionally upload client log to s3 backend
sync: pull: conditionally upload content to s3 backend
api: reader: fetch chunks based on datastore backend
datastore: local chunk reader: read chunks based on backend
verify worker: add datastore backed to verify worker
verify: implement chunk verification for stores with s3 backend
datastore: create namespace marker in s3 backend
datastore: create/delete protected marker file on s3 storage backend
datastore: prune groups/snapshots from s3 object store backend
datastore: get and set owner for s3 store backend
datastore: implement garbage collection for s3 backend
ui: add datastore type selector and reorganize component layout
ui: add s3 client edit window for configuration create/edit
ui: add s3 client view for configuration
ui: expose the s3 client view in the navigation tree
ui: add s3 client selector and bucket field for s3 backend setup
tools: lru cache: add removed callback for evicted cache nodes
tools: async lru cache: implement insert, remove and contains methods
datastore: add local datastore cache for network attached storages
api: backup: use local datastore cache on s3 backend chunk upload
api: reader: use local datastore cache on s3 backend chunk fetching
datastore: local chunk reader: get cached chunk from local cache store
backup writer: refactor parameters into backup writer options struct
api: backup: add no-cache flag to bypass local datastore cache
api/datastore: implement refresh endpoint for stores with s3 backend
cli: add dedicated subcommand for datastore s3 refresh
ui: render s3 refresh as valid maintenance type and task description
ui: expose s3 refresh button for datastores backed by object store
datastore: conditionally upload atime marker chunk to s3 backend
bin: implement client subcommands for s3 configuration manipulation
bin: expose reuse-datastore flag for proxmox-backup-manager
datastore: mark store as in-use by setting marker on s3 backend
datastore: run s3-refresh when reusing a datastore with s3 backend
api/ui: add flag to allow overwriting in-use marker for s3 backend
docs: Add section describing how to setup s3 backed datastore
Cargo.toml | 2 +
docs/storage.rst | 73 ++
examples/upload-speed.rs | 17 +-
pbs-client/src/backup_writer.rs | 47 +-
pbs-config/Cargo.toml | 1 +
pbs-config/src/lib.rs | 1 +
pbs-config/src/s3.rs | 89 +++
pbs-datastore/Cargo.toml | 5 +
pbs-datastore/src/backup_info.rs | 63 +-
pbs-datastore/src/cached_chunk_reader.rs | 6 +-
pbs-datastore/src/chunk_store.rs | 29 +-
pbs-datastore/src/datastore.rs | 696 ++++++++++++++++--
pbs-datastore/src/dynamic_index.rs | 1 +
pbs-datastore/src/lib.rs | 5 +
pbs-datastore/src/local_chunk_reader.rs | 66 +-
.../src/local_datastore_lru_cache.rs | 180 +++++
pbs-datastore/src/s3.rs | 114 +++
pbs-tools/src/async_lru_cache.rs | 46 +-
pbs-tools/src/lru_cache.rs | 42 +-
proxmox-backup-client/src/benchmark.rs | 17 +-
proxmox-backup-client/src/main.rs | 26 +-
src/api2/admin/datastore.rs | 97 ++-
src/api2/admin/mod.rs | 2 +
src/api2/admin/s3.rs | 83 +++
src/api2/backup/environment.rs | 82 ++-
src/api2/backup/mod.rs | 131 ++--
src/api2/backup/upload_chunk.rs | 114 ++-
src/api2/config/datastore.rs | 149 +++-
src/api2/config/mod.rs | 2 +
src/api2/config/s3.rs | 307 ++++++++
src/api2/node/disks/directory.rs | 2 +-
src/api2/node/disks/zfs.rs | 2 +-
src/api2/reader/environment.rs | 12 +-
src/api2/reader/mod.rs | 62 +-
src/backup/verify.rs | 138 +++-
src/bin/proxmox-backup-manager.rs | 1 +
src/bin/proxmox_backup_manager/datastore.rs | 42 ++
src/bin/proxmox_backup_manager/mod.rs | 2 +
src/bin/proxmox_backup_manager/s3.rs | 102 +++
src/server/pull.rs | 76 +-
src/server/push.rs | 27 +-
src/server/sync.rs | 22 +-
src/server/verify_job.rs | 2 +-
www/Makefile | 3 +
www/NavigationTree.js | 6 +
www/Utils.js | 4 +
www/config/S3ClientView.js | 141 ++++
www/datastore/Content.js | 48 ++
www/form/S3ClientSelector.js | 33 +
www/window/DataStoreEdit.js | 132 +++-
www/window/MaintenanceOptions.js | 6 +-
www/window/S3ClientEdit.js | 148 ++++
52 files changed, 3151 insertions(+), 353 deletions(-)
create mode 100644 pbs-config/src/s3.rs
create mode 100644 pbs-datastore/src/local_datastore_lru_cache.rs
create mode 100644 pbs-datastore/src/s3.rs
create mode 100644 src/api2/admin/s3.rs
create mode 100644 src/api2/config/s3.rs
create mode 100644 src/bin/proxmox_backup_manager/s3.rs
create mode 100644 www/config/S3ClientView.js
create mode 100644 www/form/S3ClientSelector.js
create mode 100644 www/window/S3ClientEdit.js
Summary over all repositories:
59 files changed, 3293 insertions(+), 363 deletions(-)
--
Generated by git-murpp 0.8.1
More information about the pbs-devel
mailing list