From: Rishabh Dave <ridave@redhat.com>
Date: Wed, 29 Jan 2025 13:06:59 +0000 (+0530)
Subject: mon,cephfs: don't disallow "fs fail" when other FS/MDS has health...
X-Git-Tag: v20.0.0~19^2~3
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=0437952c43d9ba4844895741a15f9d56f55899cf;p=ceph.git

mon,cephfs: don't disallow "fs fail" when other FS/MDS has health...

warnings.

Fixes: https://tracker.ceph.com/issues/69707
Signed-off-by: Rishabh Dave <ridave@redhat.com>
---

diff --git a/src/mon/FSCommands.cc b/src/mon/FSCommands.cc
index 539c7315b4cc..7f122577a95c 100644
--- a/src/mon/FSCommands.cc
+++ b/src/mon/FSCommands.cc
@@ -108,11 +108,16 @@ class FailHandler : public FileSystemCommandHandler
       return -ENOENT;
     }
 
+    vector<mds_gid_t> mds_gids_to_fail;
+    for (const auto& p : fsp->get_mds_map().get_mds_info()) {
+      mds_gids_to_fail.push_back(p.first);
+    }
+
     bool confirm = false;
     cmd_getval(cmdmap, "yes_i_really_mean_it", confirm);
     if (!confirm &&
 	mon->mdsmon()->has_health_warnings({
-	  MDS_HEALTH_TRIM, MDS_HEALTH_CACHE_OVERSIZED})) {
+	  MDS_HEALTH_TRIM, MDS_HEALTH_CACHE_OVERSIZED}, mds_gids_to_fail)) {
       ss << errmsg_for_unhealthy_mds;
       return -EPERM;
     }
@@ -122,11 +127,6 @@ class FailHandler : public FileSystemCommandHandler
     };
     fsmap.modify_filesystem(fsp->get_fscid(), std::move(f));
 
-    vector<mds_gid_t> mds_gids_to_fail;
-    for (const auto& p : fsp->get_mds_map().get_mds_info()) {
-      mds_gids_to_fail.push_back(p.first);
-    }
-
     for (const auto& gid : mds_gids_to_fail) {
       mon->mdsmon()->fail_mds_gid(fsmap, gid);
     }
diff --git a/src/mon/MDSMonitor.cc b/src/mon/MDSMonitor.cc
index ba621bc6fb50..897a27653ba8 100644
--- a/src/mon/MDSMonitor.cc
+++ b/src/mon/MDSMonitor.cc
@@ -1548,7 +1548,7 @@ out:
   }
 }
 
-bool MDSMonitor::has_health_warnings(vector<mds_metric_t> warnings, mds_gid_t gid)
+bool MDSMonitor::has_health_warnings(vector<mds_metric_t> warnings, const mds_gid_t& gid)
 {
   for (auto& [_gid, health] : pending_daemon_health) {
     if (gid != MDS_GID_NONE) {
@@ -1571,6 +1571,17 @@ bool MDSMonitor::has_health_warnings(vector<mds_metric_t> warnings, mds_gid_t gi
   return false;
 }
 
+bool MDSMonitor::has_health_warnings(const vector<mds_metric_t>& warnings,
+				     const std::vector<mds_gid_t>& gids)
+{
+  for (auto& gid : gids) {
+    if (has_health_warnings(warnings, gid)) {
+      return true;
+    }
+  }
+  return false;
+}
+
 bool MDSMonitor::has_any_health_warning()
 {
   return std::any_of(
diff --git a/src/mon/MDSMonitor.h b/src/mon/MDSMonitor.h
index 2f667abef8e4..0ba5a2c32dd2 100644
--- a/src/mon/MDSMonitor.h
+++ b/src/mon/MDSMonitor.h
@@ -53,7 +53,9 @@ class MDSMonitor : public PaxosService, public PaxosFSMap, protected CommandHand
   bool prepare_update(MonOpRequestRef op) override;
   bool should_propose(double& delay) override;
   bool has_health_warnings(std::vector<mds_metric_t> warnings,
-			   mds_gid_t gid=MDS_GID_NONE);
+			   const mds_gid_t& gid=MDS_GID_NONE);
+  bool has_health_warnings(const std::vector<mds_metric_t>& warnings,
+			   const std::vector<mds_gid_t>& gids);
   bool has_any_health_warning();
 
   bool should_print_status() const {