Al-assad opened a new pull request, #70821:
URL: https://github.com/apache/doris/pull/70821

   ### Background
   
   Standalone Doris runtime images package FE, BE and Meta Service/Recycler for 
container deployments. Their entrypoints prepare configuration, including files 
mounted through Kubernetes ConfigMaps, before starting the packaged processes.
   
   ### What problem does this PR solve?
   
   Issue Number: None
   Related PR: None
   PR Type: Enhancement
   
   Problem Summary:
   
   The runtime images previously started Doris as root, preventing deployments 
that require a non-root operating-system user. Run FE, BE and MS/Recycler as a 
dedicated user and prepare writable image directories and configuration at 
build/startup time. This covers both integrated and disaggregated deployments; 
existing root-owned volumes require separate permission preparation before 
upgrading.
   
   ### Details
   
   - **Change map:** 122 additions / 33 deletions in production Dockerfiles and 
scripts, 121 added test lines, and 71 added documentation lines; no generated 
files. Dockerfiles establish identity and ownership, entrypoints prepare 
writable configuration, and the build helper forwards identity overrides.
   - **Before and after:** FE, BE and MS/Recycler now default to numeric 
UID/GID 10001:10001. ConfigMap-backed primary configuration is copied to a 
writable file before modification. MS preserves its configured FDB endpoint 
when no environment override is supplied and handles repeated startup without 
accumulating endpoint settings.
   - **Configs and defaults:** DORIS_UID and DORIS_GID build arguments default 
to 10001; docker-build.sh forwards matching environment variables. The image 
provides the Kerberos configuration symlink at build time. Kubernetes 
configuration and volume permissions must use matching identities.
   - **Impact on existing users:** Mounted storage must be writable by the 
configured identity. This PR does not migrate existing root-owned PVCs. Without 
delegated writable CPU cgroups, Workload Group CPU weights and hard limits are 
unavailable. Host kernel tuning must be supplied separately when the Operator 
system init container is skipped.
   - **Behavior change:** The default operating-system user changes from root 
to doris for standalone FE/BE/MS images. No engine, Operator, storage format or 
protocol changes are included.
   
   ### Testing
   
   - Bash syntax checks, targeted ShellCheck and git diff --check passed.
   - docker/runtime/test/non-root-config-test.sh passed under UID 10001 with 
all capabilities dropped and no-new-privileges enabled. Coverage includes 
read-only projected ConfigMap files, repeated startup, absent endpoint 
overrides and absent ConfigMaps. Removing the writable-file chmod reproduced 
Permission denied.
   - Built linux/amd64 FE, BE and MS images using the official Apache Doris 
4.1.4 binary package, whose checksum was verified. The current branch's engine 
was not rebuilt.
   - Validated fresh DorisCluster and DorisDisaggregatedCluster deployments on 
native x86_64 K3s with unmodified Doris Operator 26.0.1 and fresh local-path 
PVCs.
   - All Doris processes ran as UID/GID 10001 with CapEff=0, NoNewPrivs=1 and 
seccomp enabled. No privileged or init containers were used for Doris; Workload 
Group cgroup setup was disabled.
   - Both deployments passed SQL INSERT, Stream Load and exact five-row query 
checks. CR-driven recreation changed every Doris Pod UID while retaining PVC 
identities and all five rows. MS and Recycler both started successfully.
   - Observed transient duplicate BE registration retries and DNS/gRPC recovery 
delays after recreation; final query checks passed. These observations were not 
established to be caused by non-root execution, and their code paths are 
unchanged.
   - Not covered: existing root-owned PVC migration, other storage drivers, 
ARM64, current-master engine compatibility, WG CPU isolation, stress/HA 
testing, or recreation of the external FDB/object-storage services.
   
   ### Release note
   
   Standalone FE, BE and Meta Service/Recycler Docker images run as UID/GID 
10001:10001 by default, with configurable build-time IDs. Prepare permissions 
on existing mounted storage before upgrading. Deployments without delegated CPU 
cgroups cannot use Workload Group CPU weights or hard limits; node-level kernel 
settings must be configured externally.
   
   ### Check List (For Author)
   
   - Behavior changed: Yes (see Details).
   - Does this need documentation: Yes; runtime README includes identity, 
Operator configuration, storage permissions and limitations.
   - Pre-submission status: Partially verified: fresh linux/amd64 deployments 
with 4.1.4 binaries passed; limitations are listed in Testing.
   
   ### AI Disclosure
   
   - Tool: OpenAI Codex.
   - Generated: Runtime Dockerfile/entrypoint changes, tests, documentation and 
this PR description were prepared with AI assistance.
   - Human review: Pending maintainer review.
   
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to