pt-query-digest is coming up with no results and "0 total, 1 unique". Any idea how to solve that?

Question

It appears for some of our logs, the percona toolkit pt-query-digest tool works fine, but for others we get the following output:

# Files: /.../mysqld_slow.log
# Overall: 0 total, 1 unique, 0 QPS, 0x concurrency ______________________
# Attribute          total     min     max     avg     95%  stddev  median
# ============     ======= ======= ======= ======= ======= ======= =======
# Query size        18.19M  18.19M  18.19M  18.19M  18.19M       0  18.19M

# Profile
# Rank       Query ID   Response time Calls      R/Call     Apdx V/M   Ite
# ========== ========== ========== ========== ========== ==== ===== ======
$

Anyone know what could be wrong with my log file? it appears to be valid, having the following first 10 lines:

Sep 28 00:00:37 gcdb-master mysqld_slow_log: SELECT  `companies`.* FROM `companies`  WHERE `companies`.`id` = 576 LIMIT 1;
Sep 28 00:00:37 gcdb-master mysqld_slow_log: # User@Host: db_one[db_one] @ ip-127.0.0.1.ec2.internal [127.0.0.1]
Sep 28 00:00:37 gcdb-master mysqld_slow_log: # Query_time: 0.041188  Lock_time: 0.000046 Rows_sent: 1  Rows_examined: 46418
Sep 28 00:00:37 gcdb-master mysqld_slow_log: SET timestamp=1348790434;
Sep 28 00:00:37 gcdb-master mysqld_slow_log: SELECT  `companies`.* FROM `companies`  WHERE `companies`.`id` = 286358 LIMIT 1;
Sep 28 00:00:37 gcdb-master mysqld_slow_log: # User@Host: db_one[db_one] @ ip-127.0.0.1.ec2.internal [127.0.0.1]
Sep 28 00:00:37 gcdb-master mysqld_slow_log: # Query_time: 0.030769  Lock_time: 0.000050 Rows_sent: 1  Rows_examined: 46583
Sep 28 00:00:37 gcdb-master mysqld_slow_log: SET timestamp=1348790434;
Sep 28 00:00:37 gcdb-master mysqld_slow_log: SELECT  `companies`.* FROM `companies`  WHERE `companies`.`id` = 286679 LIMIT 1;
Sep 28 00:00:37 gcdb-master mysqld_slow_log: # User@Host: db_one[db_one] @ ip-127.0.0.1.ec2.internal [127.0.0.1]
Sep 28 00:00:37 gcdb-master mysqld_slow_log: # Query_time: 0.594351  Lock_time: 0.000038 Rows_sent: 12  Rows_examined: 342673

Answer 1

I have run some tests with your sample output, and I suspect that your file is not valid. This file, obtained from yours by cutting the syslog-like portion of each line, and supplemented with the two missing #description-lines before the first query, appears to work

# User@Host: db_one[db_one] @ ip-127.0.0.1.ec2.internal [127.0.0.1]
# Query_time: 0.041188  Lock_time: 0.000046 Rows_sent: 1  Rows_examined: 46418
SELECT  `companies`.* FROM `companies`  WHERE `companies`.`id` = 576 LIMIT 1;
# User@Host: db_one[db_one] @ ip-127.0.0.1.ec2.internal [127.0.0.1]
# Query_time: 0.041188  Lock_time: 0.000046 Rows_sent: 1  Rows_examined: 46418
SET timestamp=1348790434;
SELECT  `companies`.* FROM `companies`  WHERE `companies`.`id` = 286358 LIMIT 1;
# User@Host: db_one[db_one] @ ip-127.0.0.1.ec2.internal [127.0.0.1]
# Query_time: 0.030769  Lock_time: 0.000050 Rows_sent: 1  Rows_examined: 46583
SET timestamp=1348790434;
SELECT  `companies`.* FROM `companies`  WHERE `companies`.`id` = 286679 LIMIT 1;
# User@Host: db_one[db_one] @ ip-127.0.0.1.ec2.internal [127.0.0.1]
# Query_time: 0.594351  Lock_time: 0.000038 Rows_sent: 12  Rows_examined: 342673

as does the same file, where I have removed the first line, so as to start with a #description line, and output:

# 240ms user time, 20ms system time, 24.59M rss, 87.74M vsz
# Current date: Fri Nov  2 22:03:02 2012
# Hostname: mintaka
# Files: orig.log
# Overall: 3 total, 1 unique, 0 QPS, 0x concurrency ______________________
# Attribute          total     min     max     avg     95%  stddev  median
# ============     ======= ======= ======= ======= ======= ======= =======
# Exec time          113ms    31ms    41ms    38ms    40ms     5ms    40ms
# Lock time          142us    46us    50us    47us    49us     2us    44us
# Rows sent              3       1       1       1       1       0       1
# Rows examine     136.15k  45.33k  45.49k  45.38k  44.45k    0.00  44.45k
# Query size           234      76      79      78   76.28    1.50   76.28

# Profile
# Rank Query ID           Response time Calls R/Call Apdx V/M   Item
# ==== ================== ============= ===== ====== ==== ===== ==========
#    1 0x0C756AF10BC44B0D 0.1131 100.0%     3 0.0377 1.00  0.00 SELECT companies

# Query 1: 0 QPS, 0x concurrency, ID 0x0C756AF10BC44B0D at byte 226 ______
# This item is included in the report because it matches --limit.
# Scores: Apdex = 1.00 [1.0]*, V/M = 0.00
# Query_time sparkline: |    ^   |
# Attribute    pct   total     min     max     avg     95%  stddev  median
# ============ === ======= ======= ======= ======= ======= ======= =======
# Count        100       3
# Exec time    100   113ms    31ms    41ms    38ms    40ms     5ms    40ms
# Lock time    100   142us    46us    50us    47us    49us     2us    44us
# Rows sent    100       3       1       1       1       1       0       1
# Rows examine 100 136.15k  45.33k  45.49k  45.38k  44.45k    0.00  44.45k
# Query size   100     234      76      79      78   76.28    1.50   76.28
# String:
# Hosts        ip-127.0.0.1.ec2.internal
# Users        db_one
# Query_time distribution
#   1us
#  10us
# 100us
#   1ms
#  10ms  ################################################################
# 100ms
#    1s
#  10s+
# Tables
#    SHOW TABLE STATUS LIKE 'companies'\G
#    SHOW CREATE TABLE `companies`\G
# EXPLAIN /*!50100 PARTITIONS*/
SELECT  `companies`.* FROM `companies`  WHERE `companies`.`id` = 286358 LIMIT 1\G

So I guess that the problem is connected with log file format and possibly rotation (e.g. the file is truncated so that the initial #description is lost).

Also, I've looked a bit into the Percona utility code. The default parser (slowlog) searches for a #Time line to get timestamp, which I find in my slow-log, but is not present in yours. This does not influence reading per se, but it might skew the results.

It also uses ";\n#" as input record separator, so the syslog format is a definite "no" for the default --type slowlog. The others also do not appear to work with syslog+slowlog format.

I've tried changing the input record separator and adding a hack to remove the syslog part of each line, and the result appears to work, but I fear it does not because the results on a known slowlog aren't consistent.

I am afraid the simplest thing to do is to chop off the beginning of the file, then cut off the beginning of each line, before feeding it to the utility:

sed -e '/.*: #/,$b' -e 'd' < slow.log \
| cut -d' ' -f6- \
| pt-query-digest -

Answer 2

It seems like the tool should never have an output like that, so you might consider filing a bug report.

The error may or may not be with your log file. You might try splitting your log file into a few smaller files and seeing if the files can be parsed -- this might at least narrow down the source of the bug. If all of the smaller files are parsed correctly, maybe it is an issue related to a lack of resources or pt-query-digest being killed for taking up too many resources.