574390 - HBase table crash_reports_index_hang_id should record *all* uuids associated with a hang

Status: RESOLVED WONTFIX

(Socorro :: General, task)

Description

[Daniel Einspanjer [:dre] [:deinspanjer]]

Currently this index table will only record the last collected instance of the different process types (i.e. default or plugin) for a given hang id.

We should change it to store all of them.  There are two ways we could do so:
1. versioning.  We could alter the table to tell it to record N versions of the values.  This would be best if we decide that duplicate ooids might be interesting but are not a completely legitimate occurance. i.e. if they really are supposed to be hang *pairs*.  
  Upsides: this is an easy schema change; it protects us from *massive* amounts of duplicates bloating the index size; and the data retrieval looks much more clean.
  Downside: is that if we receive more than N duplicate reports, the earliest ones will drop out of the index.
2. datestamping. Currently we use the column name of "ids:ooid:<process_type>".  We could change this to be something like "ids:ooid:<process_type>:<timestamp>".  Then we would record each instance of a duplicate plugin report that has a different submission timestamp.
  Upsides: record all data; no schema change necessary, only client code.
  Downsides: Possible index bloat from massive amounts of duplicates; slightly harder to retrieve the data (have to parse through the multiple columns)

Thoughts?

Comment 1

[Laura Thomson :laura]

Daniel, still wanted?

Comment 2

[Daniel Einspanjer [:dre] [:deinspanjer]]

Much less important with the implementation of ES.  We can use ES searches to pull up these various associated hang crashes by their common hang id.  I'm fine with won't fixing it and eventually having a new bug opened on the webapp site to implement said functionality when desired.

Comment 3

[Laura Thomson :laura]

OK, wontfix as requested