CA-378768: capture block schedulers

[MarkSymsCtx]

No description provided.

[bernhardkaindl]

Example to get more information in the same call:

lsblk -io type,name,sched,tran,rota,log-sec,rq-size,vendor,model,serial
TYPE NAME                                                            SCHED TRAN   ROTA LOG-SEC RQ-SIZE VENDOR   MODEL            SERIAL
disk sda                                                             cfq   sata      1     512     128 ATA      ST1000NX0423     W470MWXE
part |-sda4                                                          cfq             1     512     128                           
part |-sda2                                                          cfq             1     512     128                           
part |-sda5                                                          cfq             1     512     128                           
part |-sda3                                                          cfq             1     512     128                           
lvm  | `-VG_XenStorage--03086bb3--efa1--a74e--fbc5--16077bf518e7-MGT                 1     512     128                           
part |-sda1                                                          cfq             1     512     128                           
part `-sda6                                                          cfq             1     512     128

As in this example, I'd also add -i to use ASCII characters for the tree formatting as otherwise lsblk would use Unicode characters to format the tree if it runs inside a UTF-8 locale (like when running it from the shell) but ASCII format if running it from some XAPI plugin (in which the en_US.UTF-8 default locale of XenServer) is not set. So for a clean and more-stable ASCII output, I'd recommend adding it.

Because , are used in the arguments, they would be carried over into the generated filename by cmd_output, so it would be good to add label="lsblk" to cmd_output(), to get a file like bug-report-20230622161233/lsblk.out.

If that's not done, a file like

bug-report-20230622161233/lsblk-osched,name.out

would be created.

[MarkSymsCtx]

I started with using /sys/class/block/*/queue/scheduler and this has a number of problems

1. picks up all kernel block devices, including the tdX devices for the live, active VBDs attached to VM. Where there may be many hundred of these. Filtering to just sd* and dm-* would be possible but potentially risks missing something that could be important in triaging a customer issue.
2. The output results in one file in the bugtool for each entry in /sys/class/block/*/queue which results in a lot of files to deal with when triaging the logs whereas lsblk gives the result in one single file.
3. Due to the implementation of sysfs the stat call for reading the size of the logs will always report 4KB so you very rapidly (and erroneously) exceed the size limit for the capability, or the size limit has to be set artificially high.

[bernhardkaindl]

@mg12 Do you have a preference for just the file or a more complete set of values from lsblk, and would you need additional values from lsblk? (lsblk -h has a list)

As described in my review comment for the new cmd_output call, think this would be good:

• Use -i or --ascii to not get Unicode tree formatting in the data
• Add label=lsblk-sched" to cmd_output (or so) not get "," in the archived file name
• To retrieve more info, not just the scheduler, what about this proposal:

  lsblk -io type,name,sched,tran,rota,log-sec,rq-size,vendor,model,serial

[edwintorok]

For performance testing purposes we need the scheduler, and 'rotational' is also useful (e.g. if 'cfq' is used and it has wrongly determined that an SSD is rotational because it is hidden behind too many layers of PERC controllers, etc.).

I wouldn't mind getting the rest of the fields you suggested, except for 'serial' which might have some privacy implications and we don't really need it, so best not to explicitly capture it. 'vendor/model' can be useful to determine which drive got assigned which letter, so that is probably fine.

[mg12]

@MarkSymsCtx: what @edwintorok suggested above looks good to me, essentially @bernhardkaindl's parameters minus serial:
lsblk -io type,name,sched,tran,rota,log-sec,rq-size,vendor,model