ethereum_blobs exclude from prod

[jeff-dude]

fyi @lorenz234 @0xRobin @hildobby @MSilb7 @Jam516
tagging as history shows all involved
this has started to fail in prod on the following error:

19:54:30    Database Error in model ethereum_blobs (models/_sector/blobs/ethereum/ethereum_blobs.sql)
  TrinoUserError(type=USER_ERROR, name=GENERIC_USER_ERROR, message="Task 20240911_193845_06449_mhikh.5.1.0 is failed, due to containing long running stuck splits.", query_id=20240911_193845_06449_mhikh)

i raised to the platform team to see if it's query engine related, but looks like it's new data coming through which is breaking the query planning.
first thought revolves around:

The blobs table contains lots of fun UTF8ish data, right? And that file in particular has a very fun blob that appears to contain a single blob with 300,000 null characters

seems like lots of nulls coming through unexpectedly somewhere?

if we want this spell re-enabled, feel free to open PR to remove prod exclude tag and apply fixes as necessary (or continue the conversation on how to fix).

[Pluies]

I've managed to reproduce this on dune.com, see query 4063379.

The:

,ceil(cast(length(regexp_replace(cast(blob as varchar), '0*$', '')) - 2 as double) /2 ) AS used_blob_byte_count

Is an absolute killer on one of these blocks (not sure which!). Getting the blocks themselves, without computing used_blob_byte_count, is instant.

[belen-pruvost]

Here is an alternative for trimming empty characters in a performant, varbinary native way. You can find more data about these functions in our docs

[0xRobin]

Thank you for the detailed suggestions @Pluies, @belen-pruvost!
PR with a fix is already up. ✔️