Remove GC timeout, fix GC tests

[ghost]

Closes #3436 #2129

commit a9b8c8b
Author: Lars Gierth [email protected]
Date: Fri Dec 9 17:26:53 2016 +0100

gc: remove wonky timeout

This timeout was:
- unneccessary
- based on a metric that doesn't make sense

License: MIT
Signed-off-by: Lars Gierth <[email protected]>

commit 8f0c7ee
Author: Lars Gierth [email protected]
Date: Fri Dec 9 17:28:31 2016 +0100

gc: remove unneccessary full repo scan

GetStorageUsage() is super expensive as it involves a full repo scan,
and it's already bad enough that we have do it once.

The call that's being removed here is purely for cosmetical purposes:
printing the number of bytes freed by the GC run. Let's drop it.

License: MIT
Signed-off-by: Lars Gierth <[email protected]>

commit 49802eb
Author: Lars Gierth [email protected]
Date: Fri Dec 9 18:04:55 2016 +0100

gc: fix and re-enable sharness tests

I'm bumping up the waiting time significantly.
I prefer a slightly slower test to no test at all.

License: MIT
Signed-off-by: Lars Gierth <[email protected]>

[ghost]

Oh my, it timed out on me right away again. Is there any way to avoid the sleep? sync(1)?

[Kubuxu]

Sync won't matter as it does kernel cache to disk and problem here is go-ipfs process to kernel. I would just suggest bigger files and error margins.

[whyrusleeping]

Yeah... i remember this test being really troublesome. Let me know if you need any help @lgierth

[ghost]

If we're trying to wait for IPFS flusing its writes, why not expose the state of that in the diag command?

[Kubuxu]

I don't think it will be easy.

[whyrusleeping]

@lgierth ipfs flushes its writes... what is the problem?

[Kubuxu]

The delay it will take to flush the write is unspecified AFAIK.

[whyrusleeping]

The commands shouldnt return before things get flushed... (unless you have no-flush set)

[whyrusleeping]

@lgierth progress here?

[ghost]

This is good to go, I've split the stuff about tests out into the feat/corerepo-gc branch, since it turned into a full-blown rewrite of this GC stuff. :)

This one here now only removes the timeout context around gc.GC(), and the second repo scan after the GC run.

[ghost]

OSX failure is #3496

[whyrusleeping]

This LGTM, now we can at least run gc...