From 7e2ed4af485c1567183e9ec25a502804caaa6f4c Mon Sep 17 00:00:00 2001
From: AlexMikhalev <alex@metacortex.engineer>
Date: Fri, 7 Aug 2026 19:41:33 +0100
Subject: [PATCH] ci: harden firecracker workflow diagnostics

Improve Firecracker VM creation diagnostics by logging HTTP status/body and handling curl transport errors before JSON parsing.

Run cargo nextest directly instead of through rch exec because rch's non-compilation path loses shell quoting around nextest filter expressions.

Verification:
- workflow YAML parsed
- extracted Create VM shell passed bash -n
- simulated 201, HTTP 500, and curl transport failure paths
- independent review passed
---
 .github/workflows/ci-firecracker.yml | 36 +++++++++++++++++++++++-----
 1 file changed, 30 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/ci-firecracker.yml b/.github/workflows/ci-firecracker.yml
index 01c8cdb0d..e7c07bde7 100644
--- a/.github/workflows/ci-firecracker.yml
+++ b/.github/workflows/ci-firecracker.yml
@@ -42,10 +42,30 @@ jobs:
       - name: Create VM
         id: vm
         run: |
-          RESPONSE=$(curl -sf -X POST $FCCTL_URL/api/vms \
+          RESPONSE_FILE=$(mktemp)
+          trap 'rm -f "${RESPONSE_FILE}"' EXIT
+          if HTTP_STATUS=$(curl -sS -o "$RESPONSE_FILE" -w "%{http_code}" -X POST "$FCCTL_URL/api/vms" \
             -H 'Content-Type: application/json' \
-            -d "{\"vm_type\": \"$VM_TYPE\"}")
+            -d "{\"vm_type\": \"$VM_TYPE\"}"); then
+            :
+          else
+            CURL_EXIT=$?
+            RESPONSE=$(cat "$RESPONSE_FILE")
+            echo "fcctl create transport failed: curl exit $CURL_EXIT"
+            if [ -n "$RESPONSE" ]; then
+              echo "$RESPONSE"
+            fi
+            exit 1
+          fi
+          RESPONSE=$(cat "$RESPONSE_FILE")
+          rm -f "$RESPONSE_FILE"
+          trap - EXIT
+          echo "fcctl create status: $HTTP_STATUS"
           echo "$RESPONSE"
+          if [ "$HTTP_STATUS" -lt 200 ] || [ "$HTTP_STATUS" -ge 300 ]; then
+            echo "ERROR: fcctl-web VM create failed with HTTP $HTTP_STATUS"
+            exit 1
+          fi
           VM_ID=$(echo "$RESPONSE" | python3 -c "import sys,json; print(json.load(sys.stdin)['id'])")
           if [ -z "$VM_ID" ] || [ "$VM_ID" = "null" ]; then
             echo "ERROR: failed to parse vm id from response"
@@ -144,8 +164,8 @@ jobs:
       # bigbox; see .cargo/config.toml for the Darwin `dynamic_lookup`
       # linker workaround on local Mac dev.
       #
-      # All cargo invocations are dispatched via `rch exec --` so they
-      # share rchd's queue + slot accounting with ADF agents (see
+      # Build-oriented cargo invocations are dispatched via `rch exec --` so
+      # they share rchd's queue + slot accounting with ADF agents (see
       # .docs/adr-rch-build-queue-not-firecracker-ci.md). Fail-open: if
       # rchd is down or no slot is available, rch falls through to local
       # cargo with no behaviour change.
@@ -161,7 +181,7 @@ jobs:
       - name: Install cargo-nextest
         run: |
           if ! command -v cargo-nextest >/dev/null 2>&1; then
-            cargo install cargo-nextest --locked
+            curl -LsSf https://get.nexte.st/latest/linux | tar zxf - -C "${CARGO_HOME:-$HOME/.cargo}/bin"
           fi
           cargo nextest --version
 
@@ -169,7 +189,11 @@ jobs:
         # Only test_chat_command is skipped: it requires LLM API credentials
         # not present in CI. All other failures must be fixed at the source,
         # not skipped. nextest uses filter expressions instead of --skip.
-        run: /home/alex/.local/bin/rch exec -- cargo nextest run --workspace --profile ci -E 'not test(test_chat_command)'
+        # Do not wrap this invocation in `rch exec`: rch's non-compilation
+        # command path loses shell quoting around the filter expression and
+        # makes `/bin/sh` parse the parentheses in `test(...)`.
+        # `RUSTC_WRAPPER=sccache` above still applies to rustc invocations.
+        run: cargo nextest run --workspace --profile ci -E 'not test(test_chat_command)'
 
       - name: sccache stats
         if: always()