chore: fixing tool used to run dbmate migration checks (#1419)

* chore: fixing tool used to run dbmate migration checks

* chore: include oriole version in testdb matrix going forward

* chore: fix the nested single quote

* fix: try to increase wait time to operation

* chore: attempt to wait for orioledb loaded

Author: samrose

.github/workflows/test.yml

@@ -26,7 +26,7 @@ jobs:
       - name: Set PostgreSQL versions
         id: set-versions
         run: |
-          VERSIONS=$(nix run nixpkgs#yq -- '.postgres_major[] | select(. != "orioledb-17")' ansible/vars.yml | nix run nixpkgs#jq -- -R -s -c 'split("\n")[:-1]')
+          VERSIONS=$(nix run nixpkgs#yq -- '.postgres_major[]' ansible/vars.yml | nix run nixpkgs#jq -- -R -s -c "split(\"\n\")[:-1]")
           echo "postgres_versions=$VERSIONS" >> $GITHUB_OUTPUT
   build:
     needs: prepare

nix/tools/dbmate-tool.sh.in

@@ -19,28 +19,31 @@ STAT_EXTENSION_SQL=@STAT_EXTENSION_SQL@
 cleanup() {
     echo "Cleaning up..."
 
-    # Kill overmind processes first
+    # Kill postgres processes first
+    if pgrep -f "postgres" >/dev/null; then
+        pkill -TERM postgres || true
+        sleep 2
+    fi
+
+    # Then kill overmind
     if [ -S "./.overmind.sock" ]; then
         overmind kill || true
         sleep 2
     fi
 
-    # Kill any remaining postgres processes
-    echo "Killing any remaining postgres processes..."
-    pkill -9 postgres || true
-    pkill -9 -f "tmux.*overmind.*postgresql" || true
-    
-    # Extra cleanup for tmux sessions
+    # Kill tmux sessions explicitly
+    pkill -f "tmux.*overmind.*postgresql" || true
     tmux ls 2>/dev/null | grep 'overmind' | cut -d: -f1 | xargs -I{} tmux kill-session -t {} || true
+
+    # Force kill any stragglers
+    pkill -9 -f "(postgres|tmux.*overmind.*postgresql)" || true
 
-    # Remove socket and Procfile
     rm -f .overmind.sock Procfile
-    
-    # Verify cleanup
-    remaining=$(ps aux | grep -E "(postgres|overmind|tmux.*postgresql)" | grep -v grep || true)
-    if [ ! -z "$remaining" ]; then
-        echo "Warning: Some processes might still be running:"
-        echo "$remaining"
+
+    # Final verification
+    if ps aux | grep -E "(postgres|overmind|tmux.*postgresql)" | grep -v grep >/dev/null; then
+        ps aux | grep -E "(postgres|overmind|tmux.*postgresql)" | grep -v grep
+        return 1
     fi
 }
 
@@ -143,6 +146,24 @@ wait_for_postgres() {
     return 1
 }
 
+check_orioledb_ready() {
+    local max_attempts=30
+    local attempt=1
+    
+    while [ $attempt -le $max_attempts ]; do
+        if "${PSQLBIN}/psql" -v ON_ERROR_STOP=1 -U "$PGSQL_SUPERUSER" -p "$PORTNO" -h localhost -d postgres -c "SELECT * FROM pg_am WHERE amname = 'orioledb'" | grep -q orioledb; then
+            echo "Orioledb extension is ready!"
+            return 0
+        fi
+        echo "Waiting for orioledb to be ready (attempt $attempt/$max_attempts)..."
+        sleep 2
+        attempt=$((attempt + 1))
+    done
+    
+    echo "Orioledb failed to initialize after $max_attempts attempts"
+    return 1
+}
+
 trim_schema() {
     case "$CURRENT_SYSTEM" in
     "x86_64-darwin"|"aarch64-darwin")
@@ -165,7 +186,7 @@ EOF
     while [ $count -lt $max_wait ]; do
         if [ -S "./.overmind.sock" ]; then
             # Found the socket, give it a moment to be ready
-            sleep 2
+            sleep 5
             echo "Socket file found and ready"
             break
         fi
@@ -174,6 +195,25 @@ EOF
         count=$((count + 1))
     done
 }
+perform_dump() {
+    local max_attempts=3
+    local attempt=1
+    
+    while [ $attempt -le $max_attempts ]; do
+        echo "Attempting dbmate dump (attempt $attempt/$max_attempts)"
+        
+        if dbmate dump; then
+            return 0
+        fi
+        
+        echo "Dump attempt $attempt failed, waiting before retry..."
+        sleep 5
+        attempt=$((attempt + 1))
+    done
+    
+    echo "All dump attempts failed"
+    return 1
+}
 migrate_version() {
     echo "PSQL_VERSION: $PSQL_VERSION"
     overmind kill || true
@@ -193,6 +233,13 @@ migrate_version() {
         echo "Failed to connect to PostgreSQL server"
         exit 1
     fi
+    
+    if [ "$PSQL_VERSION" = "orioledb-17" ]; then
+        if ! check_orioledb_ready; then
+            echo "Failed to initialize orioledb extension"
+            exit 1
+        fi
+    fi
 
     echo "PostgreSQL server is ready"
 
@@ -233,7 +280,7 @@ EOSQL
     fi
 
     echo "Running dbmate dump with $PSQLBIN"
-    dbmate dump
+    perform_dump
 
     echo "CURRENT_SYSTEM: $CURRENT_SYSTEM"
     if [ -f "./db/schema.sql" ]; then