kudu git commit: KUDU-1048 master should show versions of tservers, version summary

[to...@apache.org]

Repository: kudu
Updated Branches:
  refs/heads/master b7d2780ff -> 11ad075a1


KUDU-1048 master should show versions of tservers, version summary

This patch adds a version summary table and a total count of
registered tablet servers to /tablet-servers.

It also fixes the display of the registration, which was printing in
red font.

Sample: https://raw.githubusercontent.com/wdberkeley/kudu-cr/master/KUDU-1048-improved.png

Change-Id: Idd203209e3d99292018801b94ec2904b6634854f
Reviewed-on: http://gerrit.cloudera.org:8080/4104
Tested-by: Kudu Jenkins
Reviewed-by: Todd Lipcon <to...@apache.org>


Project: http://git-wip-us.apache.org/repos/asf/kudu/repo
Commit: http://git-wip-us.apache.org/repos/asf/kudu/commit/11ad075a
Tree: http://git-wip-us.apache.org/repos/asf/kudu/tree/11ad075a
Diff: http://git-wip-us.apache.org/repos/asf/kudu/diff/11ad075a

Branch: refs/heads/master
Commit: 11ad075a113dc4c63e3074b58667b587e7308a06
Parents: b7d2780
Author: Will Berkeley <wd...@gmail.com>
Authored: Sat Aug 20 12:27:50 2016 -0400
Committer: Todd Lipcon <to...@apache.org>
Committed: Mon Aug 29 22:27:43 2016 +0000

----------------------------------------------------------------------
 src/kudu/master/master-path-handlers.cc | 37 +++++++++++++++++++++++-----
 src/kudu/master/ts_descriptor.cc        | 11 +++++++++
 src/kudu/master/ts_descriptor.h         |  3 +++
 src/kudu/master/ts_manager.cc           |  9 +------
 4 files changed, 46 insertions(+), 14 deletions(-)
----------------------------------------------------------------------


http://git-wip-us.apache.org/repos/asf/kudu/blob/11ad075a/src/kudu/master/master-path-handlers.cc
----------------------------------------------------------------------
diff --git a/src/kudu/master/master-path-handlers.cc b/src/kudu/master/master-path-handlers.cc
index e45ad0e..2eb6120 100644
--- a/src/kudu/master/master-path-handlers.cc
+++ b/src/kudu/master/master-path-handlers.cc
@@ -17,6 +17,7 @@
 
 #include "kudu/master/master-path-handlers.h"
 
+#include <array>
 #include <algorithm>
 #include <boost/bind.hpp>
 #include <map>
@@ -46,8 +47,10 @@
 
 namespace kudu {
 
+using std::array;
 using consensus::ConsensusStatePB;
 using consensus::RaftPeerPB;
+using std::map;
 using std::pair;
 using std::shared_ptr;
 using std::string;
@@ -66,18 +69,40 @@ void MasterPathHandlers::HandleTabletServers(const Webserver::WebRequest& req,
   master_->ts_manager()->GetAllDescriptors(&descs);
 
   *output << "<h1>Tablet Servers</h1>\n";
+  *output << Substitute("<p>There are $0 registered tablet servers.</p>", descs.size());
 
-  *output << "<table class='table table-striped'>\n";
-  *output << "  <tr><th>UUID</th><th>Time since heartbeat</th><th>Registration</th></tr>\n";
+  map<string, array<int, 2>> version_counts;
+  vector<string> tserver_rows;
   for (const std::shared_ptr<TSDescriptor>& desc : descs) {
     const string time_since_hb = StringPrintf("%.1fs", desc->TimeSinceHeartbeat().ToSeconds());
     ServerRegistrationPB reg;
     desc->GetRegistration(&reg);
-    *output << Substitute("<tr><th>$0</th><td>$1</td><td><code>$2</code></td></tr>\n",
-                          RegistrationToHtml(reg, desc->permanent_uuid()),
-                          time_since_hb,
-                          EscapeForHtmlToString(reg.ShortDebugString()));
+
+    if (desc->PresumedDead()) {
+      version_counts[reg.software_version()][1]++;
+    } else {
+      version_counts[reg.software_version()][0]++;
+    }
+    string row = Substitute("<tr><th>$0</th><td>$1</td><td><pre><code>$2</code></pre></td></tr>\n",
+                            RegistrationToHtml(reg, desc->permanent_uuid()),
+                            time_since_hb,
+                            EscapeForHtmlToString(reg.ShortDebugString()));
+    tserver_rows.push_back(row);
+  }
+
+  *output << "<h3>Version Summary</h3>";
+  *output << "<table class='table table-striped'>\n";
+  *output << "<tr><th>Version</th><th>Count (Live)</th><th>Count (Dead)</th></tr>\n";
+  for (const auto& entry : version_counts) {
+    *output << Substitute("<tr><td>$0</td><td>$1</td><td>$2</td></tr>\n",
+                          entry.first, entry.second[0], entry.second[1]);
   }
+  *output << "</table>";
+
+  *output << "<h3>Registrations</h3>";
+  *output << "<table class='table table-striped'>\n";
+  *output << "<tr><th>UUID</th><th>Time since heartbeat</th><th>Registration</th></tr>\n";
+  *output << JoinStrings(tserver_rows, "\n");
   *output << "</table>\n";
 }
 

http://git-wip-us.apache.org/repos/asf/kudu/blob/11ad075a/src/kudu/master/ts_descriptor.cc
----------------------------------------------------------------------
diff --git a/src/kudu/master/ts_descriptor.cc b/src/kudu/master/ts_descriptor.cc
index dd28cb2..f3f4633 100644
--- a/src/kudu/master/ts_descriptor.cc
+++ b/src/kudu/master/ts_descriptor.cc
@@ -28,6 +28,13 @@
 #include "kudu/master/master.pb.h"
 #include "kudu/tserver/tserver_admin.proxy.h"
 #include "kudu/util/net/net_util.h"
+#include "kudu/util/flag_tags.h"
+
+DEFINE_int32(tserver_unresponsive_timeout_ms, 60 * 1000,
+             "The period of time that a Master can go without receiving a heartbeat from a "
+             "tablet server before considering it unresponsive. Unresponsive servers are not "
+             "selected when assigning replicas during table creation or re-replication.");
+TAG_FLAG(tserver_unresponsive_timeout_ms, advanced);
 
 using std::make_shared;
 using std::shared_ptr;
@@ -136,6 +143,10 @@ MonoDelta TSDescriptor::TimeSinceHeartbeat() const {
   return now - last_heartbeat_;
 }
 
+bool TSDescriptor::PresumedDead() const {
+  return TimeSinceHeartbeat().ToMilliseconds() >= FLAGS_tserver_unresponsive_timeout_ms;
+}
+
 int64_t TSDescriptor::latest_seqno() const {
   std::lock_guard<simple_spinlock> l(lock_);
   return latest_seqno_;

http://git-wip-us.apache.org/repos/asf/kudu/blob/11ad075a/src/kudu/master/ts_descriptor.h
----------------------------------------------------------------------
diff --git a/src/kudu/master/ts_descriptor.h b/src/kudu/master/ts_descriptor.h
index 86a953e..701acd1 100644
--- a/src/kudu/master/ts_descriptor.h
+++ b/src/kudu/master/ts_descriptor.h
@@ -66,6 +66,9 @@ class TSDescriptor {
   // from this TS.
   MonoDelta TimeSinceHeartbeat() const;
 
+  // Return whether this server is presumed dead based on last heartbeat time.
+  bool PresumedDead() const;
+
   // Register this tablet server.
   Status Register(const NodeInstancePB& instance,
                   const ServerRegistrationPB& registration);

http://git-wip-us.apache.org/repos/asf/kudu/blob/11ad075a/src/kudu/master/ts_manager.cc
----------------------------------------------------------------------
diff --git a/src/kudu/master/ts_manager.cc b/src/kudu/master/ts_manager.cc
index dd95cc5..fd1b36d 100644
--- a/src/kudu/master/ts_manager.cc
+++ b/src/kudu/master/ts_manager.cc
@@ -24,13 +24,6 @@
 #include "kudu/gutil/strings/substitute.h"
 #include "kudu/master/master.pb.h"
 #include "kudu/master/ts_descriptor.h"
-#include "kudu/util/flag_tags.h"
-
-DEFINE_int32(tserver_unresponsive_timeout_ms, 60 * 1000,
-             "The period of time that a Master can go without receiving a heartbeat from a "
-             "tablet server before considering it unresponsive. Unresponsive servers are not "
-             "selected when assigning replicas during table creation or re-replication.");
-TAG_FLAG(tserver_unresponsive_timeout_ms, advanced);
 
 using std::shared_ptr;
 using std::string;
@@ -107,7 +100,7 @@ void TSManager::GetAllLiveDescriptors(vector<shared_ptr<TSDescriptor> > *descs)
   descs->reserve(servers_by_id_.size());
   for (const TSDescriptorMap::value_type& entry : servers_by_id_) {
     const shared_ptr<TSDescriptor>& ts = entry.second;
-    if (ts->TimeSinceHeartbeat().ToMilliseconds() < FLAGS_tserver_unresponsive_timeout_ms) {
+    if (!ts->PresumedDead()) {
       descs->push_back(ts);
     }
   }