Config option for removing bbl on assembly?

Config option for removing bbl on assembly?

[Andy Smith]

Hi,

Could we have a config setting for removing the bad blocks list on
assembly?

I'm aware that one could put:

CREATE bbl=no

in mdadm.conf and then newly created arrays won't have a bad blocks
list. But in many cases, system arrays are created before an
mdadm.conf exists, so they get a bbl.

So, to remove a bbl I know that you have to stop the array and
assemble it like:

# mdadm /dev/mdX --assemble --update=no-bbl

Again, for arrays with the actual system on it this is rather
inconvenient. Even having to have some downtime of the service to do
this is not great.

So how about a config option telling mdadm to remove empty bbl the
next time arrays are assembled? If there is a bbl with entries in it
then it could warn and not remove it (so no force).

Cheers,
Andy

Re: Config option for removing bbl on assembly?

[Roy Sigurd Karlsbakk]

> Could we have a config setting for removing the bad blocks list on
> assembly?
> 
> I'm aware that one could put:
> 
> CREATE bbl=no
> 
> in mdadm.conf and then newly created arrays won't have a bad blocks
> list. But in many cases, system arrays are created before an
> mdadm.conf exists, so they get a bbl.
> 
> So, to remove a bbl I know that you have to stop the array and
> assemble it like:
> 
> # mdadm /dev/mdX --assemble --update=no-bbl
> 
> Again, for arrays with the actual system on it this is rather
> inconvenient. Even having to have some downtime of the service to do
> this is not great.
> 
> So how about a config option telling mdadm to remove empty bbl the
> next time arrays are assembled? If there is a bbl with entries in it
> then it could warn and not remove it (so no force).

…or just remove the whole bbl, since it has virtually no function and didn't even have it when it was introduced 10 years ago. Even back then, drives had automatic reallocation of bad sectors and even if the docs say the bbl is built on write errors alone, this seems to not be the case. If a drive is bad, it should get thrown out, not patched up in software in the belief that this will make things better. That the bbl is replicated to new members and thus never removed, is part of this. I posted a thread about this earlier.

Vennlig hilsen

roy
-- 
Roy Sigurd Karlsbakk
(+47) 98013356
http://blogg.karlsbakk.net/
GPG Public key: http://karlsbakk.net/roysigurdkarlsbakk.pubkey.txt
--
Hið góða skaltu í stein höggva, hið illa í snjó rita.

Re: Config option for removing bbl on assembly?

[Andy Smith]

Hi Roy,

On Sat, Sep 12, 2020 at 12:58:45PM +0200, Roy Sigurd Karlsbakk wrote:
> …or just remove the whole bbl, since it has virtually no function
> and didn't even have it when it was introduced 10 years ago.

Yes, I saw your recent thread on the subject, and the others going
back years, and it did motivate me to think more about this.

I haven't seen an opinion from the current maintainers but I read that
Neil Brown thought it should not be removed but instead the bugs
should be found and fixed. So there doesn't seem to be any prospect
of this feature being removed.

Given that, and that there is a config option to disable it, I was
hoping it would be deemed less controversial to also have a way to
remove on next (incremental) assembly.

If you can't have your first choice of having the feature removed,
would you support that?

As an aside, I'm having a devil of a game trying to get the Debian
installer to let me use "CREATE bbl=no". Not only does
debian-installer use a config file at /tmp/mdadm.conf, but it
appears to constantly rewrite it every time you enter the MD
configuration section, so realistically it seems that the only way
to create arrays without a bbl is to do it manually from the d-i
shell.

    https://lists.debian.org/debian-boot/2020/09/msg00074.html

Cheers,
Andy

Re: Config option for removing bbl on assembly?

[Roy Sigurd Karlsbakk]

Will anyone look into this issue, please? It really is a problem.

Vennlig hilsen

roy
-- 
Roy Sigurd Karlsbakk
(+47) 98013356
http://blogg.karlsbakk.net/
GPG Public key: http://karlsbakk.net/roysigurdkarlsbakk.pubkey.txt
--
Hið góða skaltu í stein höggva, hið illa í snjó rita.

----- Original Message -----
> From: "Andy Smith" <andy@strugglers.net>
> To: "Linux Raid" <linux-raid@vger.kernel.org>
> Sent: Friday, 4 September, 2020 09:45:57
> Subject: Config option for removing bbl on assembly?

> Hi,
> 
> Could we have a config setting for removing the bad blocks list on
> assembly?
> 
> I'm aware that one could put:
> 
> CREATE bbl=no
> 
> in mdadm.conf and then newly created arrays won't have a bad blocks
> list. But in many cases, system arrays are created before an
> mdadm.conf exists, so they get a bbl.
> 
> So, to remove a bbl I know that you have to stop the array and
> assemble it like:
> 
> # mdadm /dev/mdX --assemble --update=no-bbl
> 
> Again, for arrays with the actual system on it this is rather
> inconvenient. Even having to have some downtime of the service to do
> this is not great.
> 
> So how about a config option telling mdadm to remove empty bbl the
> next time arrays are assembled? If there is a bbl with entries in it
> then it could warn and not remove it (so no force).
> 
> Cheers,
> Andy

Re: Config option for removing bbl on assembly?

[Andy Smith]

Hi Roy,

On Sat, Nov 28, 2020 at 09:15:00PM +0100, Roy Sigurd Karlsbakk wrote:
> Will anyone look into this issue, please? It really is a problem.

Since it seems you can remove the BBL by removing and re-adding a
device I no longer see a need for a new option that removes it on
assembly.

Ideally I remember to do the remove and re-add before the machine is
put into service, but failing that just doing a remove and re-add
one time seems reasonable enough.

Cheers,
Andy