search

ipfs-shipyard / ipfs-dns-deploy

A circleci friendly Docker image for pinning things to cluster and updating dns
https://hub.docker.com/r/ipfs/ipfs-dns-deploy/tags
MIT License
38 stars 4 forks source link

add content to collab cluster instead of local ipfs #13

Closed gmasgras closed 3 years ago

gmasgras commented 3 years ago

This should fix the CI issues where the local ipfs instance goes away before the collab cluster gets a chance to fetch the locally added content.

gmasgras commented 3 years ago

@hsanjuan @olizilla

olizilla commented 3 years ago

The naming is confusing here... I understand --local in this command to mean "do an ipfs add on host that you are targeting with this ipfs-cluster-ctl command, rather than "local to the current machine".

Regardless I think this fix of removing the flag will have the desired effect of making the command wait until the cid is fully repliated on cluster before it returns. It should be slower but less likely to appear to succeed then not.

olizilla commented 3 years ago

I will defer to @hsanjuan on this one tho.

hsanjuan commented 3 years ago

No, please keep. I don't think this fixes the issue we have seen, might make things worse.

gmasgras commented 3 years ago

No, please keep. I don't think this fixes the issue we have seen, might make things worse.

Why won't this fix the issue ? My understanding is that the cluster doesn't get a chance to fetch all of the content from the local ipfs node running in CI because it gets stopped right after adding content locally. Without the --local flag, content would be added straight to the cluster, which should fix this issue

olizilla commented 3 years ago

It may not fix the issue in cluster, but we do really really want this ipfs-dns-deploy script to fail or stall if cluster is stalled. Right now CI builds are succeeding and dnslinks are getting updated with cluster unable to provide the blocks. Better to fail in CI than in prod.

@hsanjuan my assumption here is that if we remove the --local flag, then the ci script would stall and time out if cluster failed to pin. Is that not the case?