I don't know how related is to this, but Ben and I had to spend a lot of time to make it work flawlessly, to stream a tar over a network pipe, check if you can find something useful for this https://github.com/kubernetes-sigs/kind/blob/b6bc112522651d98c81823df56b7afa511459a3b/pkg/cluster/internal/logs/logs.go#L39-L60

Yeah nice trick the shell invocation to have access to pipes...
I'm using tail to skip the first n bytes of the archive when resuming, however the best one is the pattern where I use a reader of a reader to catch errors and keep track of the transferred data to reconnect without untarAll noticing.

that is only valid if the source has not changed during the re connection, right?

yup, in this respect I don't improve on current implementation - however if your use case is to retrieve GBs of backups, usability is much higher
I might add a flag to set a max retry, but for the moment I'm waiting for some feedback on the issue with a prebuilt binary

For backwards compatibility and when not using retries (the proposed default, above) please use only tar, since tail isn't required in that case.

I'd prefer this to be opt-in for starters, so default to be 0, and don't use -t as the shorthand, since that's being used elsewhere as shorthand for --tty so that might be confusing here a bit.

For backwards compatibility and when not using retries (the proposed default, above) please use only tar, since tail isn't required in that case.