Exit with code 1 instead of aborting when the OSD fails to open its devices

Blockstore errors at startup, such as a data device that cannot be opened,
escaped from the etcd callback that creates the blockstore, and configuration
errors escaped from osd_t construction. vitastor-osd then died with
"terminate called after throwing std::runtime_error" and a core dump. Print
the error and exit with code 1, so that a startup error can be told from a
crash. Other exceptions still abort.

.gitea/workflows/test.yml

@@ -1913,6 +1913,24 @@ jobs:
           echo ""
         done
 
+  test_osd_open_error:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: /root/vitastor/tests/test_osd_open_error.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_enospc:
     runs-on: ubuntu-latest
     needs: build

src/osd/osd.cpp

@@ -236,7 +236,16 @@ void osd_t::init_blockstore(std::function<void()> on_init)
     if (!json_is_true(this->config["disable_blockstore"]))
     {
         auto bs_cfg = json_to_string_map(this->config);
-        this->bs = bs_factory(bs_cfg);
+        try
+        {
+            this->bs = bs_factory(bs_cfg);
+        }
+        catch (std::exception & e)
+        {
+            fprintf(stderr, "[OSD %ju] %s\n", this->osd_num, e.what());
+            force_stop(1);
+            return;
+        }
         // Pre-configure pool PG shards
         for (auto & pool_item: st_cli->pool_config)
         {

src/osd/osd_main.cpp

@@ -67,13 +67,21 @@ int main(int narg, char *args[])
     ring_loop_t *ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
     epoll_manager_t *epmgr = new epoll_manager_t(ringloop);
     auto st_cli = new etcd_state_client_http_t(epmgr->tfd);
-    osd = new osd_t(
-        config, ringloop, epmgr->tfd, std::unique_ptr<etcd_state_client_t>(st_cli),
-        [ringloop, tfd = epmgr->tfd](blockstore_config_t & cfg)
-        {
-            return blockstore_i::create(cfg, ringloop, tfd);
-        }
-    );
+    try
+    {
+        osd = new osd_t(
+            config, ringloop, epmgr->tfd, std::unique_ptr<etcd_state_client_t>(st_cli),
+            [ringloop, tfd = epmgr->tfd](blockstore_config_t & cfg)
+            {
+                return blockstore_i::create(cfg, ringloop, tfd);
+            }
+        );
+    }
+    catch (std::exception & e)
+    {
+        fprintf(stderr, "%s\n", e.what());
+        return 1;
+    }
     osd->start();
     while (1)
     {

tests/run_tests.sh

@@ -149,6 +149,8 @@ OLD=1 ./test_resize_auto.sh
 
 ./test_osd_tags.sh
 
+./test_osd_open_error.sh
+
 ./test_enospc.sh
 SCHEME=xor ./test_enospc.sh
 IMMEDIATE_COMMIT=1 ./test_enospc.sh

tests/test_osd_open_error.sh

@@ -0,0 +1,14 @@
+#!/bin/bash -ex
+
+# An OSD that cannot open its data device must exit with code 1, not abort with a core dump
+
+. `dirname $0`/common.sh
+
+wait_etcd
+
+code=0
+build/src/osd/vitastor-osd --osd_num 1 --data_device ./testdata/bin/missing --etcd_address $ETCD_URL &> ./testdata/osd1.log || code=$?
+grep -q 'Failed to open data device' ./testdata/osd1.log || format_error "OSD did not report the open error"
+[[ $code -eq 1 ]] || format_error "OSD exited with code $code instead of 1"
+
+format_green OK