Re: [gentoo-alt] tree verification in prefix

[Michael Weiser]

Hi Fabian,

On Thu, Feb 22, 2018 at 08:24:25AM +0100, Fabian Groffen wrote:

> Please excuse my snipping.

I was rambling, wasn't I. :) I'll snip some more for now.

> > > hashgen currently runs in 30s or so on the tree to generate
> > > manifests.
I recompiled gcc overnight and got the following numbers today on one
of the ARM SBCs:

[root@linux:~] time ./hashgen /usr/portage/
real    7m28.071s
user    10m8.355s
sys     0m57.435s
[root@linux:~] time ./hashgen /usr/portage/
real    7m6.195s
user    9m54.715s
sys     0m54.123s

[root@linux:~] time gemato create -K /var/lib/gentoo/gkeys/keyrings/gentoo/release/pubring.gpg -f -S -t -H SHA512\ BLAKE2B  /usr/portage
INFO:root:Creating Manifests in /usr/portage...
INFO:root:/usr/portage updated in 599.46 seconds

real    10m5.431s
user    8m54.606s
sys     0m59.072s

So there is a 30% speedup at least.

> Odd, was this rsync1 or rsync2?
[...]
> path.  Instead rsync1 and rsync2 are doing full generation both of them.
> To make this be equal, I had to play a trick with timestamps, I
> basically set the timestamp to the git commit time.  Maybe this plays a
> role too?

I just retried: timestamp.chk seemed to be in sync between rsync1 and
rsync2 at first: A sync from rsync1 syncs some files and later syncs
even against rsync2 find timestamp.chk matching.

When I delete timestamp.chk and retry several times until the other
server round-robins in, emaint sync rsyncs what seems like all
second-level Manifest.gzs (net-fs, dev-libs, etc.) and what seems like
the whole metadata/md5-cache directory.

When having synced from rsync2, gemato verify failed on
dev-cpp/Manifest.gz. When having synced from rsync1, it failed on
net-fs/Manifest.gz.

After waiting for the next refresh, emaint synced selective 2nd level
manifests and md5-cache from rsync2 again. It still failed verification
on dev-cpp/Manifest.gz.

On rsync1, timestamp.chk still differed after that first sync and emaint
sync again synced selective 2nd and 3rd level manifests and again what
seemed like the whole md5-cache. Verification with gemato still failed
on net-fs/Manifest.gz.

So the hash discrepancies seem to survive regeneration.

Then I deleted the whole $PORTDIR and sync from scratch. This was
against rsync1. Verification still fails on net-fs/Manifest.gz.

Deleting timestamp.chk again and running against rsync2, it behaved as
before and again failed at dev-cpp/Manifest.gz.

$ gemato  verify -K /Users/michael/b/pubring.gpg -R -j1
/usr/local/gentoo/usr/portage/
INFO:root:Manifest timestamp: 2018-02-22 19:26:44 UTC
INFO:root:Valid OpenPGP signature found:
INFO:root:- primary key: 0204A8ABD003E57A9558850DBA08091EC6317B3C
INFO:root:- subkey: 0204A8ABD003E57A9558850DBA08091EC6317B3C
INFO:root:- timestamp: 2018-02-22 19:26:44 UTC
INFO:root:Verifying /usr/local/gentoo/usr/portage/...
ERROR:root:Manifest mismatch for dev-cpp/Manifest.gz
  BLAKE2B: expected:
64d7c4bd55a14e8c7296e8185ad9654db39cfc095107411c97b5f425856f780c0b2dffcef436c07bc07c8832506943e7d80ab5eaf2923eb4bc419dea3a8d071a,
have:
6b698c9af8c1bf5012ee01ea308718b2f09330a181b48e663a27977885b75bd439a64d568d59de6a2a17bcad86cedf0cd8cda28361155c382badadc0d369843d
  SHA512: expected:
a7d12f2653817a47cc76de6850f8a9ab22bb952f2df1d1029cb23805f868b1d6610a2bc35d1f13666890ed1c9648907e25e949c78f75e2318065b400872e719f,
have:
070bb46740c8ecc565d23dcc35cedc3bb96d6014b083da8428be00e4008ef2e2d882d3b8e8047fe84e16542b090d02c881c08ce60b052de76406f65caf6dd893

Can you perhaps confirm these hash discrepancies on the servers?

> I only checked rsync1 last time, maybe rsync2 isn't as equal as I
> thought.

They both individually seem to produce hash discrepancies on different
files.
-- 
Thanks,
Michael