[logback-dev] [JIRA] Commented: (LBCLASSIC-170) DbAppender discards all exception' details

[Ceki Gulcu (JIRA)]

    [ http://jira.qos.ch/browse/LBCLASSIC-170?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=11658#action_11658 ] 

Ceki Gulcu commented on LBCLASSIC-170:
--------------------------------------

A recent commit [1] has DBAppender output the exception class and message as the first line. Root causes are also output recursively. 
The table schema remains unchanged. 

Admittedly, placing each stack trace line in a new row in string form (without explicit columns for the class, file and line number) is a little silly. Instead of aggregating the stack trace into a single line, a separate column for the various exception fields would have been better because it would allow searches based on exception classes/messages or stack traces.

IMO, the nicest approach table structure would be akin to:

CREATE TABLE logging_event_exception (
   event_id INT NOT NULL,
   i SMALLINT NOT NULL,  -- 0 for the topmost exception, incremented by one for each nested exception
   exception_class VARCHAR(256) NOT NULL,
   exception_msg  VARCHAR(1024),
   stack_trace VARCHAR(16384)
   PRIMARY KEY(event_id, i),
   FOREIGN KEY (event_id) REFERENCES logging_event(event_id));

where i would be incremented by one for each 'root cause'.

WDYT?

[1] http://github.com/ceki/logback/commit/130277bce0


> DbAppender discards all exception' details
> ------------------------------------------
>
>                 Key: LBCLASSIC-170
>                 URL: http://jira.qos.ch/browse/LBCLASSIC-170
>             Project: logback-classic
>          Issue Type: Bug
>          Components: appender
>    Affects Versions: 0.9.18
>            Reporter: Tomasz Nurkiewicz
>            Assignee: Ceki Gulcu
>             Fix For: 0.9.19
>
>
> DbAppender uses a single table logging_event_exception to persist stack traces, inserting one row per each stack frame. Also, each row has its parent event id. Unfortunately this approach loses a lot of exception details, making it unable to fully recreate ILoggingEvent instance that was logged by the DbAppender. List of attributes that are missing:
> * exception message
> * exception class name
> * exception root cause
> Also logging_event_exception saves the result of ch.qos.logback.classic.spi.StackTraceElementProxy#toString (which refers to java.lang.StackTraceElement#toString) instead of saving each element of java.lang.StackTraceElement separately. Now, in order to recreate StackTraceElement instance one must parse (probably using regular expression) string saved in trace_line attribute.
> My suggestion would be to create two tables:
> * logging_event_throwable with columns: message, class_name and cause foreign key referencing other logging_event_throwable instance
> * logging_event_stack_frame with columns: class_name, method, file, line and a foreign key referencing logging_event_throwable + indexing column
> Also logging_event would have a optional foreign key to logging_event_throwable.
> This modifications in database schema would allow to fully recreate ILoggingEvent. For backward compatibility ch.qos.logback.classic.db.DBAppender should remain untouched. The idea is to subclass it and override ch.qos.logback.classic.db.DBAppender#insertThrowable. 

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://jira.qos.ch/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira