Allow selection of IEC/JEDEC/non-SI units in podman stats memory output

[srcshelton]

Is this a BUG REPORT or FEATURE REQUEST? (leave only one on its own line)

/kind feature

Description

The output of podman stats was updated to use "human-readable" SI units (KB/MB/GB) when displaying output. This makes sense in the case of storage where SI units are the default (... arguably for marketing reasons in this case, but then GbE is nominally 1,000,000,000bps rather than 1,073,741,824bps).

However, when it comes to memory sizes, this is not particularly helpful - especially when the display then doesn't match up with the interpretation of units used by the parameters of the various --memory* options.

For example, if I use --memory 8g then podman stats shows a MEM USAGE / LIMIT output of 8.59GB... which isn't wrong, but doesn't aid the human-readability of the output compared to simply reading 8GiB!

The best non-breaking way to fix this might be to add additional format options - there's already a .MemUsage template placeholder - could .MemUsageBin/.MemUsageIEC/.MemUsageJEDEC be added to select binary measurements?

Output of podman version:

Version:      2.2.1
API Version:  2.1.0
Go Version:   go1.15.5
Git Commit:   a0d478edea7f775b7ce32f8eb1a01e75374486cb
Built:        Wed Dec  9 11:33:11 2020
OS/Arch:      linux/amd64

Have you tested with the latest version of Podman and have you checked the Podman Troubleshooting Guide?

Yes

[rhatdan]

Interested in opening a PR to add this functionality?

[srcshelton]

Interested in opening a PR to add this functionality?

I've had a bash at this - please see #8957

I also notice that, even before I changed anything, if a custom format is specified for podman stats then table headings are no longer output - even if a prefix of table is added to the format string, meaning that the actual effect of table is to use tabs rather than spaces to separate columns.

Is this intended? is the thinking that, just as sorting can be handled by post-processing the output (which I believe was the decision the project settled on w.r.t. sorted output?), so can re-adding the (now missing) headers?

(The issue being that the column-widths are unknown and so output also has to be piped through column or similar in order to line up, and it's suddenly getting rather intensive to get updating sorted container status with explanatory headings which align!)

[rhatdan]

I have no idea if this was a concious decision. Thanks for the PR.