Arch Linux

donallen

Member

Registered: 2009-11-30

Posts: 23

Kernel upgrade failure

I'm running Arch on several machines, and recently noticed (today being the most recent) that pacman -Syu fails when attempting to upgrade the linux package. The error today was a complaint about improper number of arguments or a NULL arg, something cryptic like that. I went into the mirrorlist file (yes, it's the latest one), where I have all the US sites uncommented, and I commented out the first few and repeated the upgrade attempt. It worked. My question is, why doesn't pacman, upon seeing the failure of an upgrade, try the next available mirror in this situation? I know it does that in other circumstances (e.g., a mirror is unresponsive). It seems like it ought to do so in this situation, too, rather than the user having to manually juggle mirrors.

/Don Allen

Trilby

Inspector Parrot

Registered: 2011-11-29

Posts: 30,462

Website

Re: Kernel upgrade failure

First, while this is less of an issue as you are not seeking a solution to the error, saying that an error message was "something cryptic" is beyond vague so it will be hard to have a specific discusion about the circumstances and how pacman would/should respond.  So, the following is more general.

It sounds like you probably got a corrupted package.  It could be a problem with the mirror, or possibly there was a problem during transfer.  Pacman will try the second mirror if it fails to download from the first mirror.  The issue you had was not a failed download: pacman successfully downloaded the tarball, (most likely) successfully decompressed it, but a error came later.  Pacman is wonderfully written, but it is not an A.I. system: it has no way of knowing the problem was do to a corrupted download.  Now package signing helps with this, but again without knowing what the error was or when in the installation process it occurred, I'm not going to waste my typing discussing the alternatives.  If you want more info, provide more info.

Don't manually juggle mirrors, and don't just use whatever mirror happens to be first in the default mirror list.  Check out tools like reflector.

edit: Also, did you try a second time from the first mirror?  Or did you change mirrors right away?  If it was the latter, is there any reason to think chaning the mirrors was the relevant difference and not just a coincidence?  All we know is that it failed once, and worked once.  Without the error message we don't even know what 'flavor' of fail it was.

Last edited by Trilby (2012-08-11 12:54:14)

---

"UNIX is simple and coherent" - Dennis Ritchie; "GNU's Not Unix" - Richard Stallman

donallen

Member

Registered: 2009-11-30

Posts: 23

Re: Kernel upgrade failure

First, while this is less of an issue as you are not seeking a solution to the error, saying that an error message was "something cryptic" is beyond vague so it will be hard to have a specific discusion about the circumstances and how pacman would/should respond.  So, the following is more general.

It sounds like you probably got a corrupted package.  It could be a problem with the mirror, or possibly there was a problem during transfer.  Pacman will try the second mirror if it fails to download from the first mirror.  The issue you had was not a failed download: pacman successfully downloaded the tarball, (most likely) successfully decompressed it, but a error came later.  Pacman is wonderfully written, but it is not an A.I. system: it has no way of knowing the problem was do to a corrupted download.  Now package signing helps with this, but again without knowing what the error was or when in the installation process it occurred, I'm not going to waste my typing discussing the alternatives.  If you want more info, provide more info.

Don't manually juggle mirrors, and don't just use whatever mirror happens to be first in the default mirror list.  Check out tools like reflector.

edit: Also, did you try a second time from the first mirror?  Or did you change mirrors right away?  If it was the latter, is there any reason to think chaning the mirrors was the relevant difference and not just a coincidence?  All we know is that it failed once, and worked once.  Without the error message we don't even know what 'flavor' of fail it was.

Yes, you are right that my report was vague. I was in the midst of trying to get the system in question upgraded and not thinking about bug reports or forum postings. After the dust settled, it occured to me to post what I did, and at that point the details were long gone from the screen. I did go to the pacman log, which provided no information whatsoever about the error, oddly. Thus, what I posted was from memory and vague, as you correctly observed.

I disagree with your statement about pacman and AI. Perhaps it's true that it has "no way of knowing the problem was due to a corrupted download". I think that's irrelevant. What it did know was that the attempted upgrade failed. It could then offer to retry the process with another mirror, giving the user the option to accept or reject the offer. No AI involved. "This didn't work, we have options, would you like to try one of them?"

As to your question about number of tries: I did try multiple times before editing mirrorlist, so I think it's likely there was broken-ness in the mirror that pacman was originally using.

Thanks for the pointer to 'reflector' -- I will take your advice and have a look at it.