From ee0797da6dc59c4f4115b7d2583fcba3916017f1 Mon Sep 17 00:00:00 2001 From: Christian Ebner Date: Tue, 14 May 2024 12:37:00 +0200 Subject: [PATCH v6 pxar proxmox-backup 00/65] fix #3174: improve file-level backup This series of patches implements an metadata based file change detection mechanism for improved pxar file level backup creation speed for unchanged files. The chosen approach is to split pxar archives on creation via the proxmox-backup-client into two separate data and upload streams, one exclusive for regular file payloads, the other one for the rest of the pxar archive, which is mostly metadata. On consecutive runs, the metadata archive of the previous backup run, which is limited in size and therefore rapidly accessed is used to lookup and compare the metadata for entries to encode. This assumes that the connection speed to the Proxmox Backup Server is sufficiently fast, allowing the download and chaching of the chunks for that index. Changes to regular files are detected by comparing all of the files metadata object, including mtime, acls, ecc. If no changes are detected, the previous payload index is used to lookup chunks to possibly re-use in the payload stream of the new archive. In order to reduce possible chunk fragmentation, the decision whether to reuse or reencode a file payload is deferred until enough information is gathered by adding entries to a look-ahead cache. If the padding introduced by reusing chunks falls below a threshold, the entries are referenced, the chunks are reused and injected into the pxar payload upload stream, otherwise they are discated and the files encoded regularly. Patches 13 and 14 are to be applied to the pxar repository only after patch 50 in the series, for the patches to compile in a sequential chain. The following lists the most notable changes included in this series since the version 5: - Fix an issue where the payload chunker was not correctly reset after suggested or forced boundaries. - Added regression tests for payload chunker and chunk stream. The following lists the most notable changes included in this series since the version 4: - Increase open file handle limit to hard limit and adapt lookahead cache size dynamically (thanks a lot to Thomas for pointing this out and providing the necessary background information). This helps with the reuse of multiple entries being contained within the same chunk, otherwise exceeding padding threshold and being therefore reencoded instead. - Fix payload chunker scan to only scan up until chunk pos in case a suggested boundary is chosen. - Fix issue with decoder state being not set to correct `InDirectory` after reading prelude and getting root directory entry. - Fix issue with kept back chunk injection when the chunk follows a range discontinuity. - Add regression test for pxar create with metadata archive and payload index reference. The following lists the most notable changes included in this series since the version 3: - Rework the whole reused chunk injection and accounting logic and use lockless async `mpsc::channel`s instead of `Arc>>`. - Reworked lookahead caching logic to use payload ranges and check for possible range continuation instead of looking up the reusable dynamic entries immediately in case of a reusable entry chain. This also avoids edge cases not covered in the previous version of the patch series. This current version therefore tends to reencode small files more aggressively, since they might introduce additional unwanted paddings. - Correctly cover also hardlinks for the reuse logic, avoiding to reencode these entries. - Add additional dedicatet chunker implementation for payload data stream, allowing the archiver to suggest boundaries to the chunker to reduce padding for reused chunks. - Add additional `change-detection-mode=data`, in order to allow creating split archives with fully reencoded payload data. - Add additional payload input readers for pxar accessor type implementations where needed. - Add additional consistency check in pxar encoder when dropping state or encoder instance. - CliParams was renamed to the more opaque Prelude, since the pxar archive does not care about its contents and this might be extended to store other information about the archive as well. - Add missing proxmox-file-restore for split archives and fix restore of tar/zip archives via WebUI. This is handled by the same decoder logic, and needed an updated payload input content range to read the data from the correct location in the payload data archive. - Additional refactoring to use the pxar reader helpers where possible. The following lists the most notable changes included in this series since the version 2: - many bugfixes regarding incorrect archive encoding by wrong offset generation, adding additional sanity checks and rather fail on encoding than produce an incorrectly encoded archive - different approach for deciding whether to reuse or reencode the entries. Previously, the entries have been encoded when a cached payload size threshold was reached. Now, the padding introduced by reusable chunks is tracked, and only if the padding does not exceed the set threshold, the entries are reused. This reduces the possible padding, at the cost of reencoding more entries. Also avoids to re-use chunks which have now large padding holes because of moved/removed files contained within. - added headers for metadata archive and payload file - added documentation An invocation of a backup run with this patches now is: ```bash proxmox-backup-client backup