blob: can't get blobs in arabica

[vgonkivs]

It's not possible to successfully retrieve the dah at a height 1373869 on arabica-11.

Steps to reproduce: 1) Launch celestia light node on arabica-11: celestia light run --core.ip=consensus.celestia-arabica-11.com --p2p.network=arabica 2) Wait until the node syncs to the needed height; 3) request blob: celestia blob get 1373869 0x8f8736b6ff9dc08065a6 0x9ede456e70b3c95ec779bd5b434d787911b1911b29629e88c80b5cc52736021b

Expected behavior: Blob's info is printed;

Actual behavior: The request is stuck until a timeout is reached(in case of CLI request forever)

Tested on the latest main.

NOTE: It happens only for this particular height.

[vgonkivs]

After debugging this issue with debug level log in both blob and share services, I've detected that the shrex client is constantly getting share.ErrNotFound:

I have also detected that it works fine if the request goes over ipld getter.

[Wondertan]

The other surprising thing/bug is that it works only when we disable shrex, while it should have worked with CascadeGetter swtiching to IPLDGetter after ShrexGetter has failed.

[vgonkivs]

The other surprising thing/bug is that it works only when we disable shrex, while it should have worked with CascadeGetter swtiching to IPLDGetter after ShrexGetter has failed.

It does not switch to the IPLDGetter as there is no timeout in the context, so we are stuck in shrex forever. This is a bug since we should be able to switch to another getter after N attempts if they are constantly failing. In case the user does want to have a timeout in this call then we should loop over all getters and not just one(Async getter fixes this issue too btw).

[vgonkivs]

https://github.com/celestiaorg/celestia-node/issues/2501

[Wondertan]

Seems like the bug is that we never actually set the timeout on this operation

[vgonkivs]

Yeah, so setting minTimeout fixes this.

[Wondertan]

Ok, we should still this or other issue around for ShrexGetter not working as expected

[srene]

so is there any option i can set in the light client to retrieve the blob, or do i have to wait for a new release that fixes the bug? thanks

[Wondertan]

@srene, do you face this particular issue? Generally, blobs are retrievable.

[srene]

this particular case was reported by me. we are not able to get a specific blob for a rollapp in dymension testnet, and is causing some issues to get full-nodes syncing.

[Wondertan]

Ah, I see. We can provide a quick fix branch for this issue.

[vgonkivs]

Hello, @srene. Do you need a test branch with the quick fix so you can test it?

[srene]

yes, that will help. thanks

[vgonkivs]

Hello, @srene. I made a fix and it is a part of v0.15.0 release.