How to recreate a 9.2 config to 10.1

[CanOfSocks]

I previously made a config in Apache traffic server in 9.2.5 using the apt package available on Debian Bookworm. I have recently made the switch to trying to use the docker container version of traffic server which is on 10.1. I was hoping to be able to use the http2 to origin features on 10.11 to reduce reuse connections on the high latency connection between traffic server and origin.

I managed to make a configuration in v9.2.5 that worked reliably at being able to resume downloads of large files, and also allowed for multi-threaded downloads with some download manager clients. I tried to use the records convert2yaml.py script to update my records.config to records.yaml and preserve the existing settings as closely as possible to the working v9.2 configuration in 10.1. I also use the slice plugin to create 10MB blocks of large files to help ensure parts of the large files are cached as the connection to the origin is slow and not very stable.

For some reason, I cannot get range requests/resumes to work in 10.1. When I run curl directly on the machine, I can receive range requests as expected from the origin server, and the resumable behaviour is restored when I rollback to a snapshot of the innstallation with traffic server v9.2.

I have attached a zip of the /etc/trafficserver of both the debian apt v9.2 and docker 10.11 configs for comparison.

Alternatively, if someone has a reliable configuration for partial content requests, I would love some more experienced insight of what I can improve.

Thanks

trafficserver-v9.2.zip

trafficserver-v10.1.zip

[BharathPESU]

Hi — I’d like to take this issue and help diagnose/fix the regressions in partial content / range requests between Traffic Server v9.2.5 (Debian apt) and v10.1 (Docker).

What I’ve observed so far:

Range/resume requests work when using the v9.2.5 setup, but stop working after moving to the v10.1 Docker image.

curl against the origin still returns valid range responses.

Rolling back to v9.2 restores resumable behaviour.

I used convert2yaml.py to translate records.config → records.yaml and I’m also using the slice plugin (10MB blocks) in both setups.

I’ve attached a zip with /etc/trafficserver from both environments for comparison.

Plan / first steps I will take:

Compare effective records.yaml / records.config differences and evaluate any new or changed default settings in 10.x that could affect range handling or cache partial responses.

Verify plugin and header handling differences (slice plugin, Cache-Control, Range, Accept-Ranges, conditional responses).

Reproduce the issue with minimal config and a reproducible origin that serves 206 Partial Content for range requests, to isolate whether the problem is in ATS request parsing, cache lookup/serve logic, or plugin interaction.

Test scenarios with http/1.1 vs http/2 to see if protocol differences between v9 and v10 affect behavior.

Propose a config change or code fix and open a PR or provide step-by-step config guidance depending on findings.

Could you please assign this issue to me? I’m happy to open a PR or post incremental findings here as I go.