feat(proxy): add prefixed CLI env var fallbacks

[pntech20]

Summary

• add LITELLM_* aliases for the env-backed proxy CLI options
• keep the existing unprefixed names as fallbacks so current deployments keep working
• cover the env var ordering in test_proxy_cli.py

Notes

TIMEOUT_WORKER_HEALTHCHECK was added after the issue was filed, so this updates the full current set of env-backed proxy flags rather than only the original ten.

Testing

• uv run --extra proxy --extra extra_proxy pytest tests/test_litellm/proxy/test_proxy_cli.py::TestProxyInitializationHelpers::test_run_server_prefers_litellm_prefixed_envvars -q
• uv run --extra proxy --extra extra_proxy pytest tests/test_litellm/proxy/test_proxy_cli.py -q (32 passed, 1 unrelated failure because hypercorn is not installed in this local env)

Closes #27791

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
2 out of 3 committers have signed the CLA.

✅ Cyberfilo
✅ cwang-otto
❌ pntech20
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This PR adds LITELLM_*-prefixed environment variable aliases for all 11 env-backed proxy_cli.py CLI options, keeping the existing unprefixed names as ordered fallbacks so current deployments are unaffected.

• proxy_cli.py: Each targeted @click.option has its envvar changed from a plain string to a two-element list [\"LITELLM_FOO\", \"FOO\"], giving the prefixed form priority while preserving backward compatibility.
• test_proxy_cli.py: A new test, test_run_server_prefers_litellm_prefixed_envvars, introspects Click's parameter metadata to assert the exact envvar list (including order) for all 11 updated options.

Confidence Score: 4/5

Safe to merge; changes are purely additive and backward-compatible — existing deployments using the unprefixed env vars continue to work unchanged.

The production code change is straightforward and low-risk: extending Click's envvar from a string to a list is a well-understood pattern with no side-effects on existing behavior. The only gap is that the accompanying test validates Click's stored parameter metadata rather than exercising actual env var resolution at runtime, so a Click regression in list-envvar priority would go undetected.

The test file could benefit from a runtime resolution test, but neither file requires blocking attention.

Important Files Changed

Filename	Overview
litellm/proxy/proxy_cli.py	Adds LITELLM_* prefixed env var aliases for 11 CLI options, keeping existing unprefixed names as fallbacks; fully backward-compatible.
tests/test_litellm/proxy/test_proxy_cli.py	New test verifies Click parameter metadata (envvar list and ordering) for all 11 updated options; does not exercise runtime env var resolution via CliRunner.

_{Reviews (1): Last reviewed commit: "feat(proxy): add prefixed cli env var fa..." | Re-trigger Greptile}

[greptile-apps]

Test only validates metadata, not runtime resolution priority

The test confirms that each Click parameter stores the expected envvar list in the correct order, but it never actually invokes run_server with conflicting env vars to verify that Click resolves LITELLM_HOST over HOST at runtime. If Click's list-envvar resolution behavior ever regressed (e.g., due to a version change that reversed iteration order), this test would still pass. A complementary test using CliRunner with both env vars set — one to a sentinel value — would give stronger proof of the precedence contract.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[oss-pr-review-agent-shin]

🤖 litellm-agent: Auto-merge skipped — the staging branch shin_agent_oss_staging_05_18_2026 has 2 commit(s) not in your branch. Merging as-is would produce a confusing diff on the staging PR.

Please rebase your branch onto shin_agent_oss_staging_05_18_2026 and push; the agent will re-review automatically.

[pntech20]

Rebased this branch onto shin_agent_oss_staging_05_18_2026 and pushed the updated branch.

[veria-ai]

PR overview

Medium: Pause toggle does not block model routing

This PR adds an Admin UI pause/resume switch for DB models. The new call sends blocked in a shape the existing PATCH endpoint ignores, so the dashboard reports success without making the model unavailable.

Security review

• 1 new security issue(s) were flagged in the latest review.
• 1 issue(s) remain open on this pull request.

Risk: 5/10

[veria-ai]

Medium: Pause toggle does not block routing

/model/{model_id}/update parses this as updateDeployment, which only merges model_info, model_name, and litellm_params; the top-level blocked field is ignored while the request still succeeds. A user with an existing key can keep sending requests to a model after an admin clicks Pause, so send and persist the flag through the backend contract and ensure routing rejects blocked deployments before showing success.