Redshift LISTAGG frame clause - sql

Question

I am trying to aggregate strings, but limited to only the preceding rows, not the whole partition. Does anyone know how to do this in Redshift?
What I am trying to achieve is the appended_event_namespace column below.
This is what I've tried so far.
LISTAGG(event_namespace, '/')
WITHIN GROUP (ORDER BY tstamp_true)
OVER (PARTITION BY acct_id) AS appended_event_namespace
This results in the full ApplicationLaunch/CategoryBrowse/NotificationCenter/UserProfile aggregation on every single row instead of what is in the desired screenshot.
The difficulty is in getting it to only append up to the current row since there doesn't seem to be a frame-clause for Redshift's LISTAGG(). Thanks for any ideas that may help.

Answer 1

You can hack this together with another query. Start with your appended_event_namespace as the result of your original LISTAGG
SELECT event_namespace,
SUBSTRING(appended_event_namespace,
1,
POSITION(event_namespace,appended_event_namespace) + LEN(event_namespace) - 1
) as appended_event_namespace_cum
FROM your_table;
Basically, you take your aggregated, ordered string, and then take the first N characters where N is ([where it appears in the aggregated string ]+[its length]), which will cut out everything after that item. This gives you a cumulative namespace.

Answer 2

LISTAGG with frame clause is not supported in RS yet. If you have some columns that you can use for partitioning and ordering you can make a self join (not so performant but would accomplish what you want):
SELECT
t1.id
,t2.tstamp_true
,t1.event_namespace
,LISTAGG(t2.event_namespace,'/') WITHIN GROUP (ORDER BY t2.tstamp_true)
FROM your_table t1
JOIN your_table t2
ON t1.id=t2.id
AND t1.tstamp_true>=t2.tstamp_true
GROUP BY 1,2,3
Alternatively, if you want to avoid self join you can build a JSON with the following structure using LISTAGG:
[{tstamp_true_1,event_namespace_1},{tstamp_true_N,event_namespace_N},...]
and write a Python UDF that takes such JSON for the given group of rows and tstamp_true of the given row and returns the path (the function would need to filter the tstamp_true_N values earlier than the second parameter and concatenate filtered event_namespace_N values for the output)