Frustrating emerge interruptions causing more blocks...

eccerr0r

Just a rant or somethingrather, I don't think this is really a support request, at least I'm not going to provide enough details since I already got past it :)

Anyway, I decided to upgrade my years out of date PVR box. emerge -uDNa --keep-going @world reported upwards 900 packages. That is, after I got all the blocks and masks cleared up, I finally got it to cleanly give me a update list. So I let it have at it.

A few hours later after 150 packages merged, it failed. Okay, what up. It's one of those packages I neglected to do the gcc 5.4.0 C++ ABI update. So I go ahead and emerge --oneshot that package.

Then I emerge --resume...

<FAIL>

What the ... not sure but somehow I destroyed the resume, not exactly sure how as I believe that emerge --oneshot was the only one.

Anyway I started emerge -uDNa --keep-going @world again. It fired right up and started building again.

Then an hour later... BOOM! failed again. What the heck? Well, it failed on my screen bomb (to make sure portage does not update gnu-screen and lock me out of my screen session which contains the emerge). Okay. Fine. Even with the --keep-going. Grr, it should have kept going. Whatever. I emerge --oneshot screen manually after I quit screen to ensure I don't lose my screen update session.

One more time with emerge --resume...

<FAIL> WHAT? A perl block?? Uggh... Okay, back to emerge -uDNa --keep-going @world again...

OH NO... BLOCK GALORE! And wants me to unmask perl-5.24 again!

Fortunately I was able to fix the problem by emerge -C a virtual package, no harm, no foul. Just a bit frustrating when the build order changes such that a whole new set of packages try to install when a package fails to install. I kind of get why this occurs as the currently installed packages have changed, but just wanted to vent a bit since --resume has failed me twice in this session...

900 packages pending update isn't a record for me at least... and talloc just failed again (bah, distcc problem...)
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Zucca · Posted: Mon Jun 17, 2019 7:12 pm Post subject:

Lately I've seen more and more blocks because some (installed, old*) package requires an old version of the backage causing the block but @world update would want to update it. Also without the newest version of the blocker I can't update most of the other packages.
*) which, in most cases, would be also updated.
So basically I have seen these stange blockers which aren't really blocking anything...

My solution has been to carefully emerge -C the blocker and maybe some other packages too. Then update @world.
This goes without saying, but I'll say it anyways, or else I'll get comments warning me: emerge -C is potentially dangerous. I don't use it to remove any python packages for example or any compiler or script language interpreter. Yes. It has left my system in somewhat crippled state like some gui toolkit not working, but that's minor, since it is only temporary state.

Then sometimes after I have solved the mess I keep seeing somewhat similar cases like you have. And --keep-going does not help.
Usually it's again some odd package blocker which can be --oneshotted or -C'd.
_________________
..: Zucca :..
Gentoo IRC channels reside on Libera.Chat.
--

Hu · Moderator Joined: 06 Mar 2007 Posts: 21631

Since we're already throwing around dangerous options, another possibility that can work in some cases is to use --nodeps to bypass the dependency logic, which also prevents Portage detecting blockers and USE flag constraint violations. This can be useful if you're sure that a particular upgrade order will be ultimately successful and less trouble than the path Portage wants, which may involve intermediate steps necessary to avoid temporary breakage. For example, if you want to rebuild a widely used Python library to support only a newer Python, Portage will normally insist that all consuming packages be updated at the same time due to their PYTHON_TARGETS constraints. You, as the administrator, may look at the list and decide you'd rather have the reverse dependencies broken for the duration than deal with updating in the order that Portage wants. --nodeps will allow you to do that. Of course, if a dependency was actually important, the build might fail.

eccerr0r · Posted: Tue Jun 18, 2019 3:20 am Post subject:

Yeah this was simply annoying because before I started the update, I had solved every block and dependency - if there weren't any ABI problems or files that didn't get downloaded/downloaded properly, it should have completed without a hitch without needing to manually unmerge anything!

However it stopped, and after trying to fix the outstanding problem that it stopped on, --resume didn't work ... so I had to resolve all the new blockers which theoretically shouldn't be blocking since ...well, they were solved already before I started the update!

I guess the trick is to figure out what to unmerge/nodeps...which can be tricky.

BTW, finally, the machine is done updating. I ended up having to depclean the old mythtvplugins (it isn't in portage any longer and as it's linked against the old mythtv so it wouldn't work anyway) as apparently it causes mythfrontend to fail starting up the video routines.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Tony0945 · Posted: Tue Jun 18, 2019 3:52 am Post subject:

My guess is that because you did the --oneshot, you destroyed the pending data that --resume needs.
Maybe if you did it in a different VT or xterm, it might have the old copy in the original terminal memory. Maybe not.
Best approach, it seems to me is --resume --skipfirst, which is what --keep-going implies to me anyway.

I've used --nodeps as Hu suggested, but it is a sharp knife. You could cut yourself bad.

On an old build, emerge -e #system first ca help, but big things like gcc will be compiled twice. And as gcc compiles itself three times during the build, that makes a total of six times. More if you have multiple gcc's.

eccerr0r · Posted: Tue Jun 18, 2019 4:14 am Post subject:

The resume data should be stored in one of those metadata files, I forget what it was. All pending merge invocations lock and share that file...

In theory, or at least if I remember correctly in the past, there should have been a 2 deep "stack" of failed and passed merges, which would mean one gets bumped off if the --oneshot also fails which is quite possible. However this is never 100% clear which should and shouldn't be kept... wish that it was possible to give each merge invocation a UUID or something that you could pick one to resume, however, I have still seen it fail due to recalculation based on current install status, so it still doesn't guarantee a resume...

Hmm...