From 86c2b93e5bd72fdabbaa8b95e72c60fe58776732 Mon Sep 17 00:00:00 2001
From: yiguolei <676222867@qq.com>
Date: Mon, 4 Dec 2023 16:04:27 +0800
Subject: [PATCH] [improvement](fixreplica) move to healthy replica when fix
 replica bad (#27934)

---------

Co-authored-by: yiguolei <yiguolei@gmail.com>
---
 .../org/apache/doris/planner/OlapScanNode.java  | 17 +++++++++++++++--
 .../org/apache/doris/qe/SessionVariable.java    |  9 +++++++++
 2 files changed, 24 insertions(+), 2 deletions(-)

diff --git a/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java b/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java
index 530a9a7830..84b54f7e4d 100644
--- a/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java
+++ b/fe/fe-core/src/main/java/org/apache/doris/planner/OlapScanNode.java
@@ -774,8 +774,21 @@ public class OlapScanNode extends ScanNode {
                 // sort by replica id
                 replicas.sort(Replica.ID_COMPARATOR);
                 Replica replica = replicas.get(useFixReplica >= replicas.size() ? replicas.size() - 1 : useFixReplica);
-                replicas.clear();
-                replicas.add(replica);
+                if (ConnectContext.get().getSessionVariable().fallbackOtherReplicaWhenFixedCorrupt) {
+                    Backend backend = Env.getCurrentSystemInfo().getBackend(replica.getBackendId());
+                    // If the fixed replica is bad, then not clear the replicas using random replica
+                    if (backend == null || !backend.isAlive()) {
+                        LOG.debug("backend {} not exists or is not alive for replica {}", replica.getBackendId(),
+                                replica.getId());
+                        Collections.shuffle(replicas);
+                    } else {
+                        replicas.clear();
+                        replicas.add(replica);
+                    }
+                } else {
+                    replicas.clear();
+                    replicas.add(replica);
+                }
             }
 
             final long coolDownReplicaId = tablet.getCooldownReplicaId();
diff --git a/fe/fe-core/src/main/java/org/apache/doris/qe/SessionVariable.java b/fe/fe-core/src/main/java/org/apache/doris/qe/SessionVariable.java
index 7224b82d91..a54a54c5ae 100644
--- a/fe/fe-core/src/main/java/org/apache/doris/qe/SessionVariable.java
+++ b/fe/fe-core/src/main/java/org/apache/doris/qe/SessionVariable.java
@@ -462,6 +462,10 @@ public class SessionVariable implements Serializable, Writable {
     public static final String MATERIALIZED_VIEW_REWRITE_ENABLE_CONTAIN_FOREIGN_TABLE
             = "materialized_view_rewrite_enable_contain_foreign_table";
 
+    // When set use fix replica = true, the fixed replica maybe bad, try to use the health one if
+    // this session variable is set to true.
+    public static final String FALLBACK_OTHER_REPLICA_WHEN_FIXED_CORRUPT = "fallback_other_replica_when_fixed_corrupt";
+
     public static final List<String> DEBUG_VARIABLES = ImmutableList.of(
             SKIP_DELETE_PREDICATE,
             SKIP_DELETE_BITMAP,
@@ -1442,6 +1446,11 @@ public class SessionVariable implements Serializable, Writable {
             "Set to true to enable Decimal256 type" })
     public boolean enableDecimal256 = false;
 
+    @VariableMgr.VarAttr(name = FALLBACK_OTHER_REPLICA_WHEN_FIXED_CORRUPT, needForward = true,
+            description = { "当开启use_fix_replica时遇到故障，是否漂移到其他健康的副本",
+                "use other health replica when the use_fix_replica meet error" })
+    public boolean fallbackOtherReplicaWhenFixedCorrupt = false;
+
     // If this fe is in fuzzy mode, then will use initFuzzyModeVariables to generate some variables,
     // not the default value set in the code.
     public void initFuzzyModeVariables() {