HiSilicon D05: probing for devices to install failed

[dann frazier]

The 2019-09-26 build of eoan-live-server-arm64.iso fails on a HiSilicon D05 (arm64) system with the message "Unfortunately probing for devices to install to failed. Please report a bug on Launchpad, and if possible include the contents of the /var/log/installer/directory.".

(Note to self: target system was host d05-3)

[Ryan Harper (raharper)]

2019-09-26 21:11:56,148 block-discover:96 probing failed restricted=True
Traceback (most recent call last):
  File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/controllers/filesystem.py", line 93, in _probed
    self.model.load_probe_data(storage)
  File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/models/filesystem.py", line 1280, in load_probe_data
    self.reset()
  File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/models/filesystem.py", line 1116, in reset
    config = storage_config.extract_storage_config(self._probe_data)
  File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 1128, in extract_storage_config
    parser = pname(probe_data)
  File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 473, in __init__
    super(BcacheParser, self).__init__(probe_data)
  File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 392, in __init__
    self.probe_data_key)
ValueError: probe_data missing bcache data

Why doesn't the arm64 kernel/image have bcache module loaded?

[dann frazier (dannf)]

On Thu, Sep 26, 2019 at 4:00 PM Ryan Harper <email address hidden> wrote:
>
> 2019-09-26 21:11:56,148 block-discover:96 probing failed restricted=True
> Traceback (most recent call last):
> File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/controllers/filesystem.py", line 93, in _probed
> self.model.load_probe_data(storage)
> File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/models/filesystem.py", line 1280, in load_probe_data
> self.reset()
> File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/models/filesystem.py", line 1116, in reset
> config = storage_config.extract_storage_config(self._probe_data)
> File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 1128, in extract_storage_config
> parser = pname(probe_data)
> File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 473, in __init__
> super(BcacheParser, self).__init__(probe_data)
> File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 392, in __init__
> self.probe_data_key)
> ValueError: probe_data missing bcache data
>
>
> Why doesn't the arm64 kernel/image have bcache module loaded?

I went to retest to check this, but my next attempt succeeded. As I
was just choosing defaults, I don't think I did anything different.
Perhaps I hit an unlucky race?

[dann frazier (dannf)]

On Thu, Sep 26, 2019 at 4:56 PM dann frazier <email address hidden> wrote:
>
> On Thu, Sep 26, 2019 at 4:00 PM Ryan Harper <email address hidden> wrote:
> >
> > 2019-09-26 21:11:56,148 block-discover:96 probing failed restricted=True
> > Traceback (most recent call last):
> > File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/controllers/filesystem.py", line 93, in _probed
> > self.model.load_probe_data(storage)
> > File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/models/filesystem.py", line 1280, in load_probe_data
> > self.reset()
> > File "/snap/subiquity/1196/lib/python3.6/site-packages/subiquity/models/filesystem.py", line 1116, in reset
> > config = storage_config.extract_storage_config(self._probe_data)
> > File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 1128, in extract_storage_config
> > parser = pname(probe_data)
> > File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 473, in __init__
> > super(BcacheParser, self).__init__(probe_data)
> > File "/snap/subiquity/1196/lib/python3.6/site-packages/curtin/storage_config.py", line 392, in __init__
> > self.probe_data_key)
> > ValueError: probe_data missing bcache data
> >
> >
> > Why doesn't the arm64 kernel/image have bcache module loaded?
>
> I went to retest to check this, but my next attempt succeeded. As I
> was just choosing defaults, I don't think I did anything different.
> Perhaps I hit an unlucky race?

Ah, I've a theory on what happened. I had burned the 20190926*.1* ISO
to a USB stick, and while I retested by booting from the 20190926 ISO
(BMC Virtual CD), I'm guessing it switched to the .1 media after
initial boot. I dd'd over the USB stick and retested and the issue
came back. So, maybe something in .1 fixed this?

 -dann

[Paride Legovini (paride)]

Mmh, so it always fails with 20190926, and it always succeeds with 20190926.1. Strange. If the problem is not reproducible with the newer images we're probably not going to spend a lot of time here, but the case is definitely curious.

I know that file corruption is improbable in this case, but maybe it's worth checking. Does the hash of your 20190926 ISO image match those in [1]?

Thanks!

Paride

[1] http://cdimage.ubuntu.com/ubuntu-server/daily/20190926/

[dann frazier (dannf)]

On Fri, Sep 27, 2019 at 7:30 AM Paride Legovini
<email address hidden> wrote:
>
> Mmh, so it always fails with 20190926, and it always succeeds with
> 20190926.1. Strange. If the problem is not reproducible with the newer
> images we're probably not going to spend a lot of time here, but the
> case is definitely curious.
>
> I know that file corruption is improbable in this case, but maybe it's
> worth checking. Does the hash of your 20190926 ISO image match those in
> [1]?

Yeah, it matches. However, I did notice some squashfs errors in dmesg
(see attached image).

[Ryan Harper (raharper)]

I'm marking the curtin task invalid; it's likely that the subiquity task is invalid as well; it appears to be something with the live image build. If you find out that there's something wrong with curtin itself, please re-open the curtin task and mark it New.

[Michael Hudson-Doyle (mwhudson)]

Assuming possibly incorrectly that this is fixed now. Please reopen if not!