Backport functionality from v5 branch.

lib/logstash/outputs/jdbc.rb

@@ -6,10 +6,33 @@ require "java"
 require "logstash-output-jdbc_jars"
 require "logstash-output-jdbc_ring-buffer"
 
+# Write events to a SQL engine, using JDBC.
+#
+# It is upto the user of the plugin to correctly configure the plugin. This
+# includes correctly crafting the SQL statement, and matching the number of
+# parameters correctly.
 class LogStash::Outputs::Jdbc < LogStash::Outputs::Base
   # Adds buffer support
   include Stud::Buffer
 
+  STRFTIME_FMT = '%Y-%m-%d %T.%L'.freeze
+
+  RETRYABLE_SQLSTATE_CLASSES = [
+    # Classes of retryable SQLSTATE codes
+    # Not all in the class will be retryable. However, this is the best that 
+    # we've got right now.
+    # If a custom state code is required, set it in retry_sql_states.
+    '08', # Connection Exception
+    '24', # Invalid Cursor State (Maybe retry-able in some circumstances)
+    '25', # Invalid Transaction State 
+    '40', # Transaction Rollback 
+    '53', # Insufficient Resources
+    '54', # Program Limit Exceeded (MAYBE)
+    '55', # Object Not In Prerequisite State
+    '57', # Operator Intervention
+    '58', # System Error
+  ].freeze
+
   config_name "jdbc"
 
   # Driver class - Reintroduced for https://github.com/theangryangel/logstash-output-jdbc/issues/26
@@ -196,8 +219,15 @@ class LogStash::Outputs::Jdbc < LogStash::Outputs::Base
   def safe_flush(events, teardown=false)
     connection = nil
     statement = nil
+    
     begin
       connection = @pool.getConnection()
+    rescue => e
+      log_jdbc_exception(e)
+      raise
+    end
+
+    begin
       statement = connection.prepareStatement(@statement[0])
 
       events.each do |event|
@@ -213,6 +243,9 @@ class LogStash::Outputs::Jdbc < LogStash::Outputs::Base
       @exceptions_tracker << nil
     rescue => e
       log_jdbc_exception(e)
+      if retry_exception?(e)
+        raise
+      end
     ensure
       statement.close() unless statement.nil?
       connection.close() unless connection.nil?
@@ -224,7 +257,12 @@ class LogStash::Outputs::Jdbc < LogStash::Outputs::Base
     statement = nil
     begin
       connection = @pool.getConnection()
+    rescue => e
+      log_jdbc_exception(e)
+      raise
+    end
 
+    begin
       events.each do |event|
         next if event.cancelled?
 
@@ -240,6 +278,9 @@ class LogStash::Outputs::Jdbc < LogStash::Outputs::Base
       end
     rescue => e
       log_jdbc_exception(e)
+      if retry_exception?(e)
+        raise
+      end
     ensure
       statement.close() unless statement.nil?
       connection.close() unless connection.nil?
@@ -250,11 +291,17 @@ class LogStash::Outputs::Jdbc < LogStash::Outputs::Base
     @statement[1..-1].each_with_index do |i, idx|
       case event[i]
       when Time
-        # Most reliable solution, cross JDBC driver
-        statement.setString(idx + 1, event[i].iso8601())
+        # See LogStash::Timestamp, below, for the why behind strftime.
+        statement.setString(idx + 1, event[i].strftime(STRFTIME_FMT))
       when LogStash::Timestamp
-        # Most reliable solution, cross JDBC driver
-        statement.setString(idx + 1, event[i].to_iso8601())
+        # XXX: Using setString as opposed to setTimestamp, because setTimestamp
+        # doesn't behave correctly in some drivers (Known: sqlite)
+        #
+        # Additionally this does not use `to_iso8601`, since some SQL databases
+        # choke on the 'T' in the string (Known: Derby).
+        #
+        # strftime appears to be the most reliable across drivers.
+        statement.setString(idx + 1, event[i].time.strftime(STRFTIME_FMT))
       when Fixnum, Integer
         statement.setInt(idx + 1, event[i])
       when Float
@@ -285,4 +332,9 @@ class LogStash::Outputs::Jdbc < LogStash::Outputs::Base
       break if current_exception == nil
     end
   end
+
+  def retry_exception?(exception)
+    return (exception.class != java.sql.SQLException or 
+      RETRYABLE_SQLSTATE_CLASSES.include?(e.getSQLState[0,2]))
+  end
 end # class LogStash::Outputs::jdbc