Backend blocks are missing compression\encoding

[edgarkz]

Describe the bug
Per documentation, both the backend and wal blocks should have configured default compression, however listing a block stats using tempo cli shows no compression at all.

To Reproduce
Steps to reproduce the behavior:

1. Run cli command go run ./cmd/tempo-cli list block single-tenant xxxxxx --backend=s3
2. Result shows no compression
   ID : 18a2572b-cdea-40d8-b468-9845ccc4e95f Version : vParquet3 Total Objects : 91933 Data Size : 198 MB Encoding : none Level : 1 Window : 475701 Start : 2024-04-07 21:29:59 +0000 UTC End : 2024-04-07 21:49:52 +0000 UTC Duration : 19m53s Age : 71h57m28s

Expected behavior
Blocks should have zstd compressed by default.

Environment:
eks 1.26
tempo 2.4.1 deployed via distributed helm chart

[joe-elliott]

The compression field is used for the old v2 format. the vParquet* formats compress columns individually.

You can review the schema to see how each column is handled differently:

https://github.com/grafana/tempo/blob/main/tempodb/encoding/vparquet3/schema.go#L111

[edgarkz]

Hi @joe-elliott thanks for quick feedback,
So the snappy encoding is default and hardcoded as per schema or its a configurable option like in v2 format?
I'll raise a pr for docs to make it clear those changes are relevant only for v2.

[joe-elliott]

The snappy encoding is hardcoded. When we initially built the schema we ran a series of tests to determine which encodings performed best for which columns and they are now hardcoded in the schema. If we wanted to change a compression type we would need to cut a new parquet version.

I'll raise a pr for docs to make it clear those changes are relevant only for v2.

awesome! thank you