Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
-
None
Description
1. Functional Test Area : Replication with Leader Hard Failure (1 Topic, 3 Partitions)
2. Testcases failed :
0151 (Sync Producer, Acks = -1, No Compression)
0152 (Async Producer, Acks = -1, No Compression)
0155 (Sync Producer, Acks = -1, Compressed)
0156 (Async Producer, Acks = -1, Compressed)
3. Sample test results :
2012-10-25 18:22:20,206 - INFO - ======================================================
2012-10-25 18:22:20,206 - INFO - validating data matched
2012-10-25 18:22:20,206 - INFO - ======================================================
2012-10-25 18:22:20,206 - DEBUG - request-num-acks [-1] (kafka_system_test_utils)
2012-10-25 18:22:20,228 - INFO - no. of unique messages on topic [test_1] sent from publisher : 900 (kafka_system_test_utils)
2012-10-25 18:22:20,235 - INFO - no. of unique messages on topic [test_1] at simple_consumer_1.log : 853 (kafka_system_test_utils)
2012-10-25 18:22:20,242 - INFO - no. of unique messages on topic [test_1] at simple_consumer_2.log : 853 (kafka_system_test_utils)
2012-10-25 18:22:20,247 - INFO - no. of unique messages on topic [test_1] at simple_consumer_3.log : 853 (kafka_system_test_utils)
4. Investigations :
a. Merge log segment files per partition:
Under test_1351181987/testcase_0151/logs/broker-1/kafka_server_1_logs:
cat test_1-0/00000000000000000000.log >> merged_test_1_0/00000000000000000000.log
cat test_1-0/00000000000000000197.log >> merged_test_1_0/00000000000000000000.log
. . .
b. Retrieve all CRC from merged data log segment:
bin/kafka-run-class.sh kafka.tools.DumpLogSegments merged_test_1_0/00000000000000000000.log | grep crc | sed 's/.* crc: //' | sort -u > test_1_0_crc.log
. . .
c. Merge the CRC files together:
cat test_1_0_crc.log >> all_crc.log
cat test_1_1_crc.log >> all_crc.log
cat test_1_2_crc.log >> all_crc.log
d. Sort the merged CRC file:
cat all_crc.log | sort -u > all_crc_sorted.log
e. Get the no. of 'failed to send' CRC in producer_performance.log (70 in this case):
grep 'failed to send' producer_performance.log | sed 's/.* crc = //' | sed 's/, key = null.*//' | sort -u | wc -l
70
f. Match those 'failed to send' CRC from producer_performance.log to see how many messages eventually got retried to send successfully:
$ for i in `grep 'failed to send' ../../producer_performance-4/producer_performance.log | sed 's/.* crc = //' | sed 's/, key = null.*//' | sort -u`; do echo -n "$i => "; grep $i all_crc_sorted.log || echo "n/a"; done;
. . .
1302684126 => n/a
1456125554 => 1456125554
15299643 => n/a
1653550869 => 1653550869
1741661084 => n/a
1764395211 => 1764395211
. . .
(23 msgs are sent successfully in retry)
g. As a result, (70 messages 'failed to send' in producer_performance.log - 23 messages successfully sent in retry) = 47 messages are lost (which matches the data loss count in the test result)
Therefore, if the no. of retry is increased to a higher value, all the messages could be sent successfully.