From 1bb52ea807dbbd2f457009a08a710d0c38feca93 Mon Sep 17 00:00:00 2001
From: "Huiliang.liu" <huliu@pivotal.io>
Date: Fri, 28 Feb 2020 17:13:33 +0800
Subject: [PATCH] Add max_retries flag for gpload (#9606)

Add max_retries flag for gpload. It indicates the max times on connecting to GPDB timed out.
max_retries default value is 0 which means no retry.
If max_retries is -1 or other negative value, it means retry forever.

Test has been done manually.

( cherry pick from master commit: b891b85ba075c0be22c6215f82923e6d950c062c)
---
 gpMgmt/bin/gpload.py | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/gpMgmt/bin/gpload.py b/gpMgmt/bin/gpload.py
index 7f18cd01aa..e88e1fc6a7 100755
--- a/gpMgmt/bin/gpload.py
+++ b/gpMgmt/bin/gpload.py
@@ -18,6 +18,7 @@ Options:
     -l logfile: log output to logfile
     --no_auto_trans: do not wrap gpload in transaction
     --gpfdist_timeout timeout: gpfdist timeout value
+    --max_retries retry_times: max retry times on gpdb connection timed out. 0 means disabled, -1 means forever
     --version: print version number and exit
     -?: help
 '''
@@ -1157,6 +1158,7 @@ class gpload:
         self.startTimestamp = time.time()
         self.error_table = False
         self.gpdb_version = ""
+        self.options.max_retries = 0
         seenv = False
         seenq = False
 
@@ -1217,6 +1219,9 @@ class gpload:
                     elif argv[0]=='-f':
                         configFilename = argv[1]
                         argv = argv[2:]
+                    elif argv[0]=='--max_retries':
+                        self.options.max_retries = int(argv[1])
+                        argv = argv[2:]
                     elif argv[0]=='--no_auto_trans':
                         self.options.no_auto_trans = True
                         argv = argv[1:]
@@ -1841,6 +1846,18 @@ class gpload:
                 if recurse > 10:
                     self.log(self.ERROR, "too many login attempt failures")
                 self.setup_connection(recurse)
+            elif errorMessage.find("Connection timed out") != -1 and self.options.max_retries != 0:
+                recurse += 1
+                if self.options.max_retries > 0:
+                    if recurse > self.options.max_retries: # retry failed
+                        self.log(self.ERROR, "could not connect to database after retry %d times, " \
+                            "error message:\n %s" % (recurse-1, errorMessage))
+                    else:
+                        self.log(self.INFO, "retry to connect to database, %d of %d times" % (recurse,
+                            self.options.max_retries))
+                else: # max_retries < 0, retry forever
+                    self.log(self.INFO, "retry to connect to database.")
+                self.setup_connection(recurse)
             else:
                 self.log(self.ERROR, "could not connect to database: %s. Is " \
                     "the Greenplum Database running on port %i?" % (errorMessage,
-- 
GitLab