Commit graph

99 commits

Author SHA1 Message Date
da4533c2d3 uniquify urls in respect to sha256 checksums (#21)
this reduces 19139 urls down to 18563 urls
2024-11-14 19:25:14 +01:00
6a0ae2bcab take mirrors into account (#13) and allow upstream-caches (#5)
This is done by introducing a set of alternative download locations.
2024-11-14 17:40:22 +01:00
c376a4b70e checked: do on file-by-file basis, incrementally 2024-11-14 16:58:20 +01:00
4481923ade revise startup (as proposed in #18):
- recover git (from disk or download)
- make index.tar.gz
- start web service
- check disk (unless skip-verify-sha256)
- dump git state
- start downloads
- enable update job, and hook
2024-11-14 16:36:10 +01:00
7e09f08767 print git last modified as well as HTTP date 2024-11-14 16:35:56 +01:00
6dab71a9ac introduce more data on status page:
- remaining downloads
- identified urls
- last git fetch & fetch status

see #22
2024-11-14 15:50:26 +01:00
e51550aedc if there's a http error, return the error 2024-11-13 20:03:51 +01:00
982a35a5b1 status: sort time as newest first 2024-11-13 14:14:32 +01:00
97f68a85e9 cope with git-kv change in 0.0.5: digest returns the raw digest 2024-11-13 12:47:36 +01:00
68d2ec8e98 use git-kv 0.0.5+ API for to/of_octets 2024-11-08 15:31:50 +01:00
1f9e3e6e23 log less, now that status is around 2024-11-08 13:51:33 +01:00
c3d5c74075 use K.remote, not the entire repo 2024-11-06 13:15:10 +01:00
be87d19797 reset errors, and sort failures 2024-11-06 13:08:28 +01:00
37008e81f3 organize failures into sections 2024-11-06 12:57:43 +01:00
8ba4cfae00 group download failures by error 2024-11-04 18:43:50 +01:00
9c50538877 record and preserve opam file parsing issues 2024-11-04 18:43:21 +01:00
a47193f147 tweaks 2024-11-04 17:33:50 +01:00
1e35bfefbd initialization: potentially rename bad data 2024-11-04 17:33:33 +01:00
a9b8f18192 stream git contents, also make the tarball and the find_urls in one go 2024-11-04 17:17:52 +01:00
2312092e42 first write to a temporary filename, and rename later 2024-11-04 17:17:37 +01:00
4bec3bfbd8 restore checksum failure error 2024-11-04 16:50:42 +01:00
f48cc19fc4 drop superfluous 'unknown' 2024-11-04 16:46:01 +01:00
7689397ac3 remove bad archive list 2024-11-04 16:14:09 +01:00
e59f02a16f always use swap, remove the pending / to_delete stuff 2024-11-04 16:13:36 +01:00
456340562d Use swapfs 2024-11-01 14:35:08 +01:00
53af2665fa record free bytes in tar archive 2024-10-31 17:45:34 +01:00
f40083692a tweaks 2024-10-31 17:17:34 +01:00
eb95821b2e Add a download status page, recording:
- number of archives on disk
- current downloads
- failed downloads
2024-10-31 11:30:52 +01:00
9bfde48f43 also handle extra-source, fixes #8 2024-10-24 15:03:35 +02:00
9a2576c423 extract the opam parsing function to a separate module (opam_file.ml / Opam_file)
prepares for handling extra-source as well
2024-10-24 14:45:31 +02:00
8af9f04dd0 Hex-encode computed checksum in log message
Oops!
2024-10-18 10:43:09 +02:00
5accfcfc08 More informative log message 2024-10-18 10:26:43 +02:00
625081abff Merge pull request 'Improve logging of bad checksums' (#7) from log-bad-checksum into main
Reviewed-on: #7
2024-10-17 16:12:30 +00:00
5938a60289 Improve logging of bad checksums
We now print the computed hash \o/
2024-10-17 17:26:00 +02:00
7a71e095da Demote log level for unavailable packages
Many packages are marked unavailable these days in preparation of
opam-repository cleanup \o/ so let's be a little less noisy.
2024-10-17 17:18:16 +02:00
26643fbcde Add a message on successful --initialize-disk 2024-10-16 11:06:48 +02:00
719b4ea45d Repo tarball: use level Ustar 2024-10-16 11:05:27 +02:00
ceb4674ec2 Add code for formatting a disk 2024-10-16 11:05:24 +02:00
d36a0714e7 Use GPTar table
We expect the disk to be formatted already.
2024-10-16 10:56:20 +02:00
58656926e3 update to mirage 4.8 2024-10-11 13:31:48 +02:00
6490801ce1 Update dependencies, refactor
- Tar.3.0.0 was released! So we can drop the pin and depend on it
  directly.
- Use digestif in favor of the bygone mirage-crypto hash
- Move most of the hashing logic into archive_checksum.ml
- When checking the checksum we should check the checksum chosen, too. I
  believe this was a bug that was hidden due to opam packages always
  using one of md5, sha256 or sha512 for checksums.
2024-10-03 14:37:37 +02:00
8fc8f1c62d Update the unikernel with the new version of tar 2024-07-31 14:54:26 +02:00
c2ffbdb891 Update to mirage.4.6.0 2024-07-15 12:22:57 +02:00
1ebf370a4f compiles now 2023-05-02 16:13:43 +02:00
Robur
fd8ce3be03 more wip 2023-05-02 12:03:25 +00:00
ebeadf69d8 Skip unavailable packages 2023-02-08 16:30:56 +01:00
adf6564385 Print error and exit if we can't get commit id 2023-02-08 16:30:30 +01:00
8c7e71127f Try to remove files before marking for deletion
You know, it might just work...
2023-02-08 16:29:55 +01:00
1a6c6b8f9d Return error on bad checksum when finalizing write 2023-01-31 10:27:25 +01:00
8e326ecbc5 WIP partial writes
If possible, downloads are streamed to disk in the /pending/ directory
in the tar filesystem. If the download is successful and the checksums
match the file is renamed to its sha256 hash. Otherwise, it is moved
under /to-delete/ so it can be deleted by an operator.

Before downloading we check if it has been downloaded before, but we
need to check as well if it is being downloaded (in /pending/) or if we
unsuccessfully downloaded it before (e.g. failed checksum, stored in
/to-delete/).

It is not very elegant code, and it could do with a thorough review or
rewrite.
2023-01-25 11:34:31 +01:00