From c44c522688e3329b5881ee2baea0d5be777d5ebc Mon Sep 17 00:00:00 2001
From: Jake Smith <jake.smith@lexisnexisrisk.com>
Date: Fri, 25 Oct 2024 18:40:48 +0100
Subject: [PATCH] HPCC-32890 Add cleanjobqueues to daliadmin

cleanjobqueues scans all /JobQueues and the Clients within them,
and deletes any which are not in use (connected==0).

NB: this should only be necessary on old (pre 9.4 systems).
(see HPCC-31807 in 9.4+)

Signed-off-by: Jake Smith <jake.smith@lexisnexisrisk.com>
---
 dali/daliadmin/daadmin.cpp   | 39 ++++++++++++++++++++++++++++++++++++
 dali/daliadmin/daadmin.hpp   |  1 +
 dali/daliadmin/daliadmin.cpp |  6 ++++++
 3 files changed, 46 insertions(+)

diff --git a/dali/daliadmin/daadmin.cpp b/dali/daliadmin/daadmin.cpp
index d1ea8bd9b02..e143850e3eb 100644
--- a/dali/daliadmin/daadmin.cpp
+++ b/dali/daliadmin/daadmin.cpp
@@ -3385,4 +3385,43 @@ void removeOrphanedGlobalVariables(bool dryrun, bool reconstruct)
     }
 }
 
+void cleanJobQueues(bool dryRun)
+{
+    Owned<IRemoteConnection> conn = querySDS().connect("/JobQueues", myProcessSession(), 0, SDS_LOCK_TIMEOUT);
+    if (!conn)
+    {
+        WARNLOG("Failed to connect to /JobQueues");
+        return;
+    }
+    Owned<IPropertyTreeIterator> queueIter = conn->queryRoot()->getElements("Queue");
+    ForEach(*queueIter)
+    {
+        IPropertyTree &queue = queueIter->query();
+        const char *name = queue.queryProp("@name");
+        if (isEmptyString(name)) // should not be blank, but guard
+            continue;
+        PROGLOG("Processing queue: %s", name);
+        VStringBuffer queuePath("/JobQueues/Queue[@name=\"%s\"]", name);
+        Owned<IRemoteConnection> queueConn = querySDS().connect(queuePath, myProcessSession(), RTM_LOCK_WRITE, SDS_LOCK_TIMEOUT);
+        IPropertyTree *queueRoot = queueConn->queryRoot();
+
+        Owned<IPropertyTreeIterator> clientIter = queueRoot->getElements("Client");
+        std::vector<IPropertyTree *> toRemove;
+        ForEach (*clientIter)
+        {
+            IPropertyTree &client = clientIter->query();
+            if (client.getPropInt("@connected") == 0)
+                toRemove.push_back(&client);
+        }
+        if (!dryRun)
+        {
+            for (auto &client: toRemove)
+                queue.removeTree(client);
+        }
+        PROGLOG("Job queue '%s': %s %u stale client entries", name, dryRun ? "dryrun, there are" : "removed", (unsigned)toRemove.size());
+        queueConn->commit();
+        queueConn.clear();
+    }
+}
+
 } // namespace daadmin
diff --git a/dali/daliadmin/daadmin.hpp b/dali/daliadmin/daadmin.hpp
index 687d0d882b2..f0141760e03 100644
--- a/dali/daliadmin/daadmin.hpp
+++ b/dali/daliadmin/daadmin.hpp
@@ -95,5 +95,6 @@ extern DALIADMIN_API void daliping(const char *dalis, unsigned connecttime, unsi
 
 extern DALIADMIN_API void validateStore(bool fix, bool deleteFiles, bool verbose);
 extern DALIADMIN_API void removeOrphanedGlobalVariables(bool dryrun, bool reconstruct);
+extern DALIADMIN_API void cleanJobQueues(bool dryRun);
 
 } // namespace daadmin
\ No newline at end of file
diff --git a/dali/daliadmin/daliadmin.cpp b/dali/daliadmin/daliadmin.cpp
index 0491b7d42f1..4c56836aac6 100644
--- a/dali/daliadmin/daliadmin.cpp
+++ b/dali/daliadmin/daliadmin.cpp
@@ -93,6 +93,7 @@ void usage(const char *exe)
   printf("Other dali server and misc commands:\n");
   printf("  auditlog <fromdate> <todate> <match>\n");
   printf("  cleanglobalwuid [dryrun] [noreconstruct]\n");
+  printf("  cleanjobqueues [dryrun]\n");
   printf("  clusterlist <mask>              -- list clusters   (mask optional)\n");
   printf("  coalesce                        -- force transaction coalesce\n");
   printf("  dalilocks [ <ip-pattern> ] [ files ] -- get all locked files/xpaths\n");
@@ -576,6 +577,11 @@ int main(int argc, const char* argv[])
                         }
                         removeOrphanedGlobalVariables(dryrun, reconstruct);
                     }
+                    else if (strieq(cmd, "cleanjobqueues"))
+                    {
+                        bool dryRun = np>0 && strieq("dryrun", params.item(1));
+                        cleanJobQueues(dryRun);
+                    }
                     else if (strieq(cmd, "remotetest"))
                         remoteTest(params.item(1), true);
                     else