Lossless grouping

Follow

Lorens Kockum

• October 26, 2018 14:08

When working with Pig and Hadoop, I could aggregate data in the way below, and I am not able to reproduce this on Sumologic. Is it possible and if so how?

 

I have keys and values, simple enough:

 

Keys; Values

--------

K1; V1a

K1; V1b

K1; V1c

K2; V2a

K3; V3a

K3; V3b

K3; V3c

K3; V3d

 

As an example, using "|count by Keys" would give me

Keys; _count

-------------

K1; 3

K2; 1

K3; 4

 

I want to get

Keys;group_of_Values

-------------------

K1; {V1a; V1b; V1c}

K2; {V2a}

K3; {V3a; V3b; V3c; V3d}

 

Any way to manage this?

 

0

• 

  Piotr Woch

  • October 26, 2018 14:19

  Yes, definitively there is a way to achieve this type of grouping in Sumo by using a combination of two operators: transactionize and merge.

  Please take a look at the transactionize operator here: https://help.sumologic.com/05Search/Search-Query-Language/Transaction-Analytics/Transactionize-operator

  and some Merge examples here: https://help.sumologic.com/05Search/Search-Query-Language/Transaction-Analytics/Merge-Operator

  Hope this helps! 

  0

  Comment actions Permalink
• 

  Nick

  • October 26, 2018 15:32

  You could also use the transpose operator to accomplish this, depending on how exactly you wanted it formatted. Something like this:

  | count by keys, values
  | transpose row keys column values

  Thanks,
  Nick
  Sumo Logic - Customer Success

  0

  Comment actions Permalink
• 

  Lorens Kockum

  • December 05, 2018 11:33

  Hi, and thanks to both of you. The transactionize/merge method looks like should be able to do what I need, I *think* I've got it (but I want to state that I am disappointed by the lack of formal documentation, maybe I just haven't found it . . .)

  I start with

  | count by k, v, team, service, environment
  | fields -_count // just to make explicit that I don't need ti
  | toLong(0) as _messagetime // either transactionize or merge absolutely wants this hidden field

  If I do 

  | transactionize k (merge k, v join with ", ")

  then it looks good except that all the team/service/environment values are ignored and lost.

  If I try to correct by using
  | transactionize team, service, environment, k (merge k, v join with ", ")

  then team/service/environment are still lost and I get {group of k}, {group of v}, which is useless.

  Doing

  | transactionize k (merge k takeFirst, v join with ", ", team takeFirst, service takeFirst, environment takeFirst)

  looks like it does what I want! I still have to test with overlapping keys in different sets (for the moment I only have my test team / test service / test environment). I'm wary of the takeFirst eating values that are different.

  Again, thanks for your help.

   

  0

  Comment actions Permalink
• 

  Lorens Kockum

  • December 05, 2018 11:36

  However the Transactionize docs say that I should get transaction fields... I don't.

  0

  Comment actions Permalink

Please sign in to leave a comment.