New Flex Consumption function apps never start in our subscription — 7 dead hosts across 3 regions and 2 stacks, Premium control works (host runtime unreachable, no telemetry, no storage footprint)

Turbocharged Racing 0 Reputation points
2026-11-16T01:33:16.89+00:00

Summary

Every newly created Azure Functions Flex Consumption app in our subscription fails to start its host runtime — including minimal hello-world apps. Apps provisioned earlier in 2025 on the identical shape (same resource group, region, storage account) run normally, and an Elastic Premium (EP1) app created today as a control starts immediately. We are asking Microsoft to establish whether this is (i) a subscription-level provisioning restriction, (ii) a regional Flex Consumption infrastructure issue, or (iii) another platform-side failure — and what diagnostic surface exists when the host produces no telemetry and no storage footprint at all.

  • Subscription: c85e8e05-3c96-4a7d-8873-afb85195ba5c
  • Tenant: e57dbbb2-9548-4cac-a1ac-43b91dd6942b
  • Resource group: tcr-prod

The specimen matrix (all Flex Consumption FC1, all created 2026-11-13)

  1. ts-worker-func-staging — East US 2 — Python 3.11, managed identity — shared storage — dark (destroyed + recreated: identical)
  2. ts-worker-flex-probe — East US 2 — Python 3.11, key auth — shared storage — dark
  3. ts-worker-func-wus2 — West US 2 — Python 3.11, managed identity — fresh dedicated storage — dark
  4. probe cell A — West US 2 — Node 20 — dark
  5. probe cell B — Central US — Python 3.11 — dark
  6. ts-probe-ftp — East US 2 — Python 3.11, HTTP function — fresh storage — dark; host never started; both CLI deploy routes failed at the platform layer and the deployment container stayed empty

Control: ts-worker-func (production worker) — East US 2 — Python 3.11 — shared storage — runs normally (provisioned earlier in 2025)

Control: EP1 probe — East US 2 — Python 3.11, Elastic Premium — fresh storage — starts first try: root 200, functions enumerate, timer fires, telemetry flows

The signature (identical on every new Flex host)

  • ARM state "Running"; root URL returns HTTP 503 ("Function host is not running")
  • Functions enumeration returns "Encountered an error (InternalServerError) from host runtime"; sync triggers returns ServiceUnavailable
  • Zero Application Insights telemetry from the new hosts (connection strings verified against the working app's workspace)
  • Zero host storage footprint: no azure-webjobs-hosts container/instance, no host keys; restarts, cold stop/start, and destroy + recreate all reproduce it identically
  • Storage diagnostic logging on the shared account shows the new hosts never issue a single host-side storage operation — the only record is the deployment-side package fetch (GetBlob 200). The failure sits above the storage layer.
  • Azure Monitor metrics discriminate further: for two specimens, InstanceCount reads 1-2 and MemoryWorkingSet sits at ~459-469 MB for ~11-15 minutes, then stops — compute is assigned and a process starts, then dies in early host initialization, before any observable act. The healthy production host sustains 0.98-1.18 GB. The final read (why the process dies) sits in scale-controller logs customers cannot see.

Deployment correlation IDs (all status Success): staging twin instance 1 01e4dfcb-f115-4cac-b5ca-d4ccc1dea4df · twin instance 2 b47eb98c-1c24-4f5b-ab1d-6ea916429745 · probe 0edcfa65-6bb4-4ef3-a9a5-ce6fb18aa73c · test env graded deploy b91671a0-6fc4-4a41-9712-302aa2e2d0c0 (byte-identical package, sha256-verified).

What we have ruled out

  • Package/content: the production workload package imports and builds cleanly (remote build succeeds, deployment records Success); the hello-world package (one timer or one HTTP function, azure-functions only) fails identically. Not a stack fault (Python + Node both dark), not a regional fault (East US 2 / West US 2 / Central US all dark).
  • Storage auth form: managed identity and account-key forms both fail; the published azurerm empty-AccountKey poison string is absent.
  • RBAC/grants: identity principals hold Storage Blob Data Owner + Key Vault Secrets User, verified by-scope before and after recreate; DB principals registered with verified grant censuses.
  • App settings / Key Vault: 27/27 settings verified name- and value-level on the twin; every Key Vault reference resolves.
  • Service health: zero ServiceIssue events on the subscription in the window; the app's own detector headlines "Functions runtime is unreachable" over empty measurement tables. Resource Health is not available for Consumption-plan apps.

Question

What prevents a newly provisioned Flex Consumption host in this subscription from starting, when earlier-provisioned Flex apps run and a Premium app on the same shape starts immediately? Is there any customer-visible diagnostic surface for a host that dies in early initialization — no telemetry, no storage footprint, no host keys?

We are re-running a minimal provisioning attempt daily and will update this thread with each result (region/stack labeled) until a new Flex host starts.

Azure Functions
Azure Functions

An Azure service that provides an event-driven serverless compute platform.

0 comments No comments

6 answers

Sort by: Newest
  1. Turbocharged Racing 0 Reputation points
    2026-11-20T19:13:05.05+00:00

    Daily update — 2026-11-20

    Region: centralus

    Stack: Python 3.11, Flex Consumption FC1

    App: ts-flex-retry-120centraluspy311

    Verdict: host started — root HTTP 200, functions answers.

    Retention: KEPT-ALIVE-BREAK — the app was deliberately left alive as evidence; nothing was torn down.

    Was this answer helpful?

    0 comments No comments

  2. Turbocharged Racing 0 Reputation points
    2026-11-19T19:13:27.71+00:00

    Daily update — 2026-11-19

    Region: westus2

    Stack: Python 3.11, Flex Consumption FC1

    App: ts-flex-retry-61119westus2py311

    Result: Host started — root HTTP 200, functions answer.

    Retention: KEPT-ALIVE-BREAK — the app is deliberately left alive as evidence; nothing was torn down.

    This follows STARTED results on 2026-11-16 (northeurope) and 2026-11-18 (eastus2), and a DARK result on 2026-11-17 (westeurope).

    Was this answer helpful?

    0 comments No comments

  3. Turbocharged Racing 0 Reputation points
    2026-11-18T19:13:03.7933333+00:00

    Daily update — 2026-11-18 — eastus2 — Python 3.11 Flex Consumption (FC1)

    Attempt: 2026-11-18-eastus2-py311

    App: ts-flex-retry-61118eastus2py311

    Storage: tsflexr61118eastus2py311

    Verdict: STARTED — host started. Root HTTP 200, functions enumeration answers, /api/hello answering.

    Retention: KEPT-ALIVE-BREAK — the app and its resources were deliberately left alive as evidence; nothing was torn down.

    This is an eastus2 specimen starting today, following the 2026-11-16 northeurope STARTED result and the 2026-11-17 westeurope DARK result (root HTTP 000). The daily retry continues per the commitment in this thread.

    Was this answer helpful?

    0 comments No comments

  4. Turbocharged Racing 0 Reputation points
    2026-11-17T19:29:59.7233333+00:00

    Daily update — 2026-11-17: Region westeurope, stack Python 3.11 Flex Consumption FC1, app ts-flex-retry-17westeuropepy311, storage tsflexr17westeuropepy311. Verdict: host never started — root HTTP 000, functions not-read. Retention: specimen retained as the westeurope standing specimen for backend correlation; no superseded same-region specimen was removed (first retained in region). This follows the 2026-11-16 northeurope STARTED result — today's westeurope specimen is DARK per the provisioning script's VERDICT block.

    Was this answer helpful?

    0 comments No comments

  5. Turbocharged Racing 0 Reputation points
    2026-11-16T19:12:31.1533333+00:00

    Daily update — 2026-11-16: Region northeurope, stack Python 3.11 Flex Consumption FC1, app ts-flex-retry-6northeuropepy311. Verdict: HOST STARTED — root HTTP 200, functions enumeration answers. This is the first new Flex Consumption host to start in this subscription in this daily retry series. The app is being kept alive as evidence (not torn down). Previous daily retries and the original twin in eastus2 / other regions were dark with root 503 and host-runtime InternalServerError.

    Was this answer helpful?

    0 comments No comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.