Wait for root device inside the initrd script for ISO images

[rene]

Description

When installing EVE on devices with Virtual Media CD/DVD-ROM emulation it can take a few seconds to kernel to detect the installer media device. This is an issue when booting from an ISO because GRUB will boot kernel + initrd, but when initrd script tries to find the installer media's filesystem, the Virtual CD/DVD-ROM can be not yet available. This PR implements a retry mechanism inside the initrd script in order to wait for the root device.

[rene]

Marked as Draft for now because it needs more test...

[deitch]

So basically, rootdelay=10 really means, "if you could not find the root filesystem, wait 1 second, try again, repeat up to 10 (i.e. rootdelay) times.

That looks like it should work.

Small cleanliness nit. Would it work if, instead of the heredoc and EOF, we did:

blkid | while read -r line; do

?

[rene]

So basically, rootdelay=10 really means, "if you could not find the root filesystem, wait 1 second, try again, repeat up to 10 (i.e. rootdelay) times.

That looks like it should work.

Small cleanliness nit. Would it work if, instead of the heredoc and EOF, we did:

blkid | while read -r line; do

?

@deitch , I agree this form is cleaner but then rootdev variable will be only in the scope of the subshell executed in the pipe, so we won't get its value after the blkid | while... command, that's why the heredoc approach is used... I would also prefer the bashism form while ... done < <(blkid) but then Yetus will scream about it...

[rene]

Reworked this PR:

• No additional kernel parameters are added
• A retry mechanism is implemented inside the initrd script in order to wait for the root device. It gives up if device doesn't show up after 10 retries (one per second)

Tested with QEMU + HP device (through iLO5)

[OhmSpectator]

Fine to merge if tested =)
And I'm still curious, why rootdelay did not work on the Kernel level...

[rene]

Fine to merge if tested =) And I'm still curious, why rootdelay did not work on the Kernel level...

Because we are using ramdisk in this case (ramdisk is the initial root device), which is already available in memory (so we don't need to wait), that's what was trying to explain... the issue was happening inside the initrd script, when we try to switch root to the installer root device, that cannot be available at the moment (doesn't exist), that's why we need to wait...

[OhmSpectator]

Fine to merge if tested =) And I'm still curious, why rootdelay did not work on the Kernel level...

Because we are using ramdisk in this case (ramdisk is the initial root device), which is already available in memory (so we don't need to wait), that's what was trying to explain... the issue was happening inside the initrd script, when we try to switch root to the installer root device, that cannot be available at the moment (doesn't exist), that's why we need to wait...

Ah, got it! Thanks!

[rucoder]

LGTM

[OhmSpectator]

@rene, do we have any Tests that can check it? Or may we skip the tests?

[rene]

@rene, do we have any Tests that can check it? Or may we skip the tests?

@rene, do we have any Tests that can check it? Or may we skip the tests?

No, so we don't need to wait for Eden tests...