Chris, Thanks for your continued detailed reporting.
The short answer -- I am glad it fixes it and I will clean up the patches and queue them for the next release. I'll try to answer your detailed questions below. On Sun, Oct 15, 2017 at 01:04:36AM +0200, [email protected] wrote: > What exactly do you mean with by-hash files, perhaps the dep11_by-hash files > in /var/cache/apt-cacher/package (I had them before, see above output of > -v)? A couple of years ago APT started to include support for distributing index files named by their MD5/SHA256 hash to prevent the regular 'Hash sum mismatch' error that we used to see regularly when index files and caches were out of sync. See https://wiki.debian.org/RepositoryFormat#Acquire-By-Hash > Why do you have files which I don't have? Probably because I have multiple clients using apt-cacher with different releases. Also, in your original report you suggested you had deleted /var/cache/apt-cacher and recreated it a some point. So files that may have been cached initially would have been lost. > Do I have to configure > something special in /etc/apt/apt.conf.d/x for that? The only option I can > imagine is that I have disabled caching of packages in > /var/cache/apt/archives, as the packages are cached centrally by apt-cacher > anyway. However, I do not think this is related. Neither do I. > $ ls -la /var/cache/apt-cacher/packages > drwxr-xr-x 2 www-data www-data 4096 Okt 14 23:16 . > drwxr-xr-x 6 www-data www-data 4096 Okt 10 19:31 .. > -rw-r--r-- 1 www-data www-data 135616 Okt 14 23:16 > ftp.de.debian.org_debian_dists_testing_InRelease > -rw-r--r-- 1 www-data www-data 25494 Okt 14 23:16 > security.debian.org_dists_testing_updates_InRelease > > The following was necessary, as otherwise only the two above files were > downloaded. > > $ rm -rf /var/cache/apt/*cache.bin > $ apt-get update This seems fine: you had recently updated therefore there were no new files requested or cached. By deleting /var/cache/apt/*cache.bin you forced apt to fetch new index files through apt-cacher. > Today, as I already have installed all upgrades in the morning, I perform an > install of an arbitrary package and not a dist-uprade here. However, this > should make no difference. > > $ apt-get install aria2 [snip] > $ /tmp/apt-cacher-cleanup-double-patched.pl -v > Invoked as root, changing to www-data:www-data and re-execing. > Checking for > ftp.de.debian.org_debian_dists_testing_main_binary-amd64_Packages{,.gz,.xz,.bz2} > for > ftp.de.debian.org_debian_dists_testing_main_binary-amd64_Packages.diff_by-hash_SHA256_7caccf05e56ae4d753c02810d9f3ddff2e4b043e29dd6dae3f899e15ec66414d > Checking for > ftp.de.debian.org_debian_dists_testing_main_binary-i386_Packages{,.gz,.xz,.bz2} > for > ftp.de.debian.org_debian_dists_testing_main_binary-i386_Packages.diff_by-hash_SHA256_5b692219d1090892f18a0b1d84e5504b72574b94c3167c2bb3625fdb8330b22f > Checking for > ftp.de.debian.org_debian_dists_testing_main_i18n_Translation-en{,.gz,.xz,.bz2} > for > ftp.de.debian.org_debian_dists_testing_main_i18n_Translation-en.diff_by-hash_SHA256_210c49af092a400efe747dc414330297cd49304ce121a29fda35640d40fc4bab > Checking for > ftp.de.debian.org_debian_dists_testing_main_source_Sources{,.gz,.xz,.bz2} for > ftp.de.debian.org_debian_dists_testing_main_source_Sources.diff_by-hash_SHA256_6bfd9b104752d28149c67e633ca654b7de9d6fd567ac025425d996b7b4ad66c4 > Get /ftp.de.debian.org/debian/dists/testing/main/i18n/Translation-en.xz > Get /ftp.de.debian.org/debian/dists/testing/main/i18n/Translation-en.gz > Get /ftp.de.debian.org/debian/dists/testing/main/i18n/Translation-en.bz2 > Get /ftp.de.debian.org/debian/dists/testing/main/source/Sources.xz > Get /ftp.de.debian.org/debian/dists/testing/main/binary-amd64/Packages.xz > Get /ftp.de.debian.org/debian/dists/testing/main/binary-i386/Packages.xz > Get /ftp.de.debian.org/debian/dists/testing/InRelease These are the cruicial lines: apt-get is updating its indices using {Packages,Sources}.diff files so the complete index file doesn't get requested each time and might not be in /var/cache/apt-cacher. The second patch ensures that the full Packages and Sources files are present in /var/cache/apt-cacher for each {Packages,Sources}.diff file so that apt-cacher can correctly keep valid current packages. > Reading: ftp.de.debian.org_debian_dists_testing_main_binary-amd64_Packages.xz > Reading: ftp.de.debian.org_debian_dists_testing_main_binary-i386_Packages.xz > Reading: ftp.de.debian.org_debian_dists_testing_main_source_Sources.xz The complete index files are read > Found 158132 valid file entries and a realistic number of valid packages are found > $ ls -la /var/cache/apt-cacher/packages/ftp.de.debian.org_debian > drwxr-xr-x 2 www-data www-data 4096 Okt 14 23:34 . > drwxr-xr-x 3 www-data www-data 4096 Okt 14 23:39 .. > -rw-r--r-- 1 www-data www-data 1210410 Okt 14 23:34 aria2_1.32.0-1_amd64.deb which means this valid package is not removed. > Wow, seams that first cleaning the apt *cache.bin files to collect all need > index/hash files from the repo anew, so that apt-cacher could cache it, and > your patches did the job! > > What/why did the behavior change that makes your patches necessary, although > some > time ago all was working well? Questions over questions. I suspect removing and recreating /var/cache/apt-cacher may have exposed it. But I am happy to accept that is something apt-cacher should cope with and doesn't differ from a new install. > Now I tried it directly with the original version, just for the fun of it: > > Invoked as root, changing to www-data:www-data and re-execing. [snip] > Reading: ftp.de.debian.org_debian_dists_testing_InRelease > Reading: ftp.de.debian.org_debian_dists_testing_main_binary-amd64_Packages.xz > Reading: ftp.de.debian.org_debian_dists_testing_main_binary-i386_Packages.xz > Reading: ftp.de.debian.org_debian_dists_testing_main_source_Sources.xz [snip] > $ ls -la /var/cache/apt-cacher/packages/ftp.de.debian.org_debian > drwxr-xr-x 2 www-data www-data 4096 Okt 14 23:34 . > drwxr-xr-x 3 www-data www-data 4096 Okt 14 23:39 .. > -rw-r--r-- 1 www-data www-data 1210410 Okt 14 23:34 aria2_1.32.0-1_amd64.deb > > Now things are getting curious. Why does it now not delete the package > anymore? Maybe somehow I have all/the right index files now. Yes, the {Packages,Sources}.xz are still present. > How can I ensure that in future My second patch does that for you. You can verify , e.g., by something like a --force parameter for > apt-get update? Why should I not have had the right index files for weeks > before? A combination of reasons: because you deleted /var/cache/apt-cacher and apt-get is updating its indices with diff files and apt-cacher wasn't clever enough to realise they were missing. > Finally, just some of my thoughts on the game: How gets apt-get clean its > similar job done for the package cache in /var/cache/apt/archives? The > necessity of removing the *cache.bin files before apt-get update stems from > the fact that for the index/hash data there is always a two level caching. > When there is a hit in the inner cache (apt), the outer one is not updated > (apt-cacher). The difference was that apt has to have complete index files and therefore made sensible decisions. Until my second patch, apt-cacher just relied on what was in the cache which was determined by what clients had requested through it. I hope that clarifies. Do do some more testing and live with it for a while. If you don't mind I might pass you a cleaned up version of the patch to test? Many thanks for your help. Mark
