Add extraction progress output to GetFSFromImage/GetFSFromLayers

[mafredri]

This PR allows printing extraction progress during unpack. Since we can't access the underlying reader for the compressed data, we can only print accurate progress at every layer change. Instead we print ... every 1 second (triggered via read) to indicate that the extraction is still in progress.

Fixes coder/envbuilder#124

#2: Building stage 'mcr.microsoft.com/devcontainers/typescript-node:18-bookworm' [idx: '1', base-idx: '-1']
#2: Unpacking rootfs as cmd ADD install-vscode.sh /root/ requires it.
#2: Extracting image layers to /
#2: Extracting layer 1/22 (0.0%)
#2: Extracting layer 1/22 (0.0%) .
#2: Extracting layer 1/22 (0.0%) ..
#2: Extracting layer 2/22 (7.6%)
#2: Extracting layer 3/22 (11.3%)
#2: Extracting layer 3/22 (11.3%) .
#2: Extracting layer 3/22 (11.3%) ..
#2: Extracting layer 4/22 (21.1%)
#2: Extracting layer 4/22 (21.1%) .
#2: Extracting layer 4/22 (21.1%) ..
#2: Extracting layer 4/22 (21.1%) ...
#2: Extracting layer 4/22 (21.1%) ....
#2: Extracting layer 4/22 (21.1%) .....
#2: Extracting layer 4/22 (21.1%) ......
#2: Extracting layer 5/22 (53.4%)
#2: Extracting layer 6/22 (53.4%)
#2: Extracting layer 6/22 (53.4%) .
#2: Extracting layer 7/22 (60.4%)
#2: Extracting layer 8/22 (60.6%)
#2: Extracting layer 9/22 (60.6%)
#2: Extracting layer 10/22 (61.5%)
#2: Extracting layer 11/22 (61.5%)
#2: Extracting layer 12/22 (61.5%)
#2: Extracting layer 13/22 (61.5%)
#2: Extracting layer 14/22 (61.5%)
#2: Extracting layer 14/22 (61.5%) .
#2: Extracting layer 14/22 (61.5%) ..
#2: Extracting layer 15/22 (72.1%)
#2: Extracting layer 15/22 (72.1%) .
#2: Extracting layer 16/22 (82.0%)
#2: Extracting layer 17/22 (83.6%)
#2: Extracting layer 17/22 (83.6%) .
#2: Extracting layer 18/22 (90.5%)
#2: Extracting layer 19/22 (90.5%)
#2: Extracting layer 20/22 (90.5%)
#2: Extracting layer 21/22 (90.5%)
#2: Extracting layer 22/22 (90.5%)
#2: Extracting layer 22/22 (90.5%) .
#2: Extraction complete
#2: Using files from context: [/workspaces/vscode/.devcontainer/install-vscode.sh]

[mafredri]

Whoops, just noticed this:

#2: Extracting image layers to / (0 bytes)
#2: Extraction complete 0/0 (-9223372036854775808%)

I'll push a fix later.

[johnstcn]

Some nits, but LGTM otherwise. I don't need to review again 👍

[dannykopping]

Overall LGTM

Left a couple nits & suggestions

Q: did you consider passing a io.TeeReader to the extraction function? That might give you insight into the progress when io.Copy occurs here (I haven't validated this).

[dannykopping]

As an aside: I think it'd be even more useful to have a timestamp at the start of every log line so we could see how long things took after the fact.

[mafredri]

We could consider enabling that on the logger lever, but I don't think we should diverge for these log lines. Usually timestamps would be available e.g. via container logs or similar. Not against showing an elapsed human readable time, but then again I want to keep this simple 😄.

[mtojek]

👍

[mafredri]

Q: did you consider passing a io.TeeReader to the extraction function? That might give you insight into the progress when io.Copy occurs here (I haven't validated this).

I'm not following what this would achieve? We're already monitoring reads via the printing reader and writes we can't know anything about.

[dannykopping]

Q: did you consider passing a io.TeeReader to the extraction function? That might give you insight into the progress when io.Copy occurs here (I haven't validated this).

I'm not following what this would achieve? We're already monitoring reads via the printing reader and writes we can't know anything about.

In func (r *printAfterReader) Read(p []byte) (n int, err error) { you're printing . while the extracting is in progress, but I'm saying you could monitoring the extraction of each layer itself (I think). In any case, this is great as it is - just an idea for later if we have some big layers which take ages to extract and we need more fine-grained insight into that process.

[mafredri]

In func (r *printAfterReader) Read(p []byte) (n int, err error) { you're printing . while the extracting is in progress, but I'm saying you could monitoring the extraction of each layer itself (I think). In any case, this is great as it is - just an idea for later if we have some big layers which take ages to extract and we need more fine-grained insight into that process.

@dannykopping Ah, I see. This is something I wanted to do, but as I mentioned in the PR description, it's not possible. We only know the compressed size and the number of uncompressed bytes read. 😔 Correlating these will always be inaccurate.