-
Bug
-
Resolution: Fixed
-
Major
-
None
-
Timestamper 1.8.8
Jenkins 1.651.3
Viewing the consoleText log with timestamps works, but viewing the plaintext log with time stamps does not work properly. The log is cut short, and the time stamps ends too early.
Console text has proper time stamps:
00:40:41.937 Finished: SUCCESS
Plain text (.../timestamps?appendLog) does not:
00:40:41.937 make[3]: Entering directory '/buildroot/otp-OTP-19.2/erts/lib_src' make[3]: Nothing to be done for 'all'.
The number of timestamps is about 16k, but there are 26k lines in the log. I can see that some parts of the log, e.g. what looks like empty lines, is handled differently in the consoleText part compared to the plaintext part:
consoleText skips empty lines:
00:00:16.744 Sending build context to Docker daemon 19.46 kB 00:00:16.834 Step 1 : FROM openjdk:8u131-jre-alpine
/timestamps does not:
00:00:16.744 Sending build context to Docker daemon 19.46 kB 00:00:16.834 00:00:16.834 00:00:16.835 Step 1 : FROM openjdk:8u131-jre-alpine
Here the time stamp becomes skewed.
It looks like the code which presents the plaintext output does not match the code for the consoleText in some cases which skews the times. This makes the plain text timestamp output essentially useless for most logs.
[JENKINS-46420] Raw log with timestamps is cut short
Darragh Bailey
added a comment - Tested the following fix and it appears to work as expected for me: https://github.com/jenkinsci/timestamper-plugin/pull/21
You can check locally by using a simple echo statement in code that uses '\r' instead of '\n' for newlines.
Raw output and console matched.
Darragh Bailey
added a comment - Tested the following fix and it appears to work as expected for me: https://github.com/jenkinsci/timestamper-plugin/pull/21
You can check locally by using a simple echo statement in code that uses '\r' instead of '\n' for newlines.
Raw output and console matched. Code changed in jenkins
User: Darragh Bailey
Path:
src/main/java/hudson/plugins/timestamper/io/LogFileReader.java
http://jenkins-ci.org/commit/timestamper-plugin/996c35a1e49f3ed914b32e105b16e57b66cbff15
Log:
JENKINS-46420 Match console end of line delimiter
Console output defaults to a limited set of end of line delimiters,
such as '\n'. Change to using the Scanner class which allows for
control of the delimiter character to ensure the LogFileReader class
matches between lines and recorded timestamps even when output contains
other newline characters.
Code changed in jenkins
User: StevenGBrown
Path:
src/main/java/hudson/plugins/timestamper/io/LogFileReader.java
http://jenkins-ci.org/commit/timestamper-plugin/43c84084e01fae21c3e825a5d0e23722bec8f556
Log:
Merge pull request #21 from electrofelix/limit-newline-characters
JENKINS-46420 Match console end of line delimiter
Compare: https://github.com/jenkinsci/timestamper-plugin/compare/2fed9af88ade...43c84084e01f
Steven G Brown
added a comment - Thanks. I've released your fix with Timestamper 1.8.9. 
Looks like in the console it is possible for log messages to contain a new line while being written out as a single entry, but the timestamper reader is assuming each new line delimits a new log entry. Consequently, appear to run out of timestamps.
Looking in a raw log file on disk, it looks like '^M' character is written out to a console log in to indicate a newline within a log line, without indicating a new log line. But the BufferedReader assumes any newline, carriage return or linefeed.
Perhaps simply switching from BufferedReader to Scanner in LogFileReader would be sufficient to solve? Allows use of a delimiter as the end of line token as described at https://stackoverflow.com/a/16712231