Multiple stats for multiple key-values with a common prefix

[mfietz]

We have log entries with multiple key-value pairs. All of the keys I'm interested in have a common prefix and all of the values are decimal numbers. Unfortunately, not every entry contains all of the keys.
An entry could look like this:
<TIMESTAMP, LOGLEVEL etc.> prefix-<SUFFIX1>=<VALUE1>, prefix-<SUFFIX3>=<VALUE2>, ...

What I want is a table where the first column contains the union of all of the keys with the common prefix, the other columns should contain statistics like median, different percentiles, standard deviation.

What I got so far:
host=... index=... | rex max_match=100 "(?<field>prefix-\w+?)=(?<val>\d+)" | stats median(val) as "Median", perc90(val) as "90th percentile", perc99(val) as "99th percentile", min(val) as "min", max(val) as "max", stdev(val) as "Standard deviation", count(val) as "count" by field

The problem: Because not all log entries contain all keys, somewhere along the pipes the value 0 seems to be assumed for those missing keys. As a result, the statistical measures get totally skewed.

I would be thankful for any help.

[somesoni2 · Accepted solution]

How about this

your base search | table prefix-* | eval temp=1 | untable temp field value | stats median(val) as "Median", perc90(val) as "90th percentile", perc99(val) as "99th percentile", min(val) as "min", max(val) as "max", stdev(val) as "Standard deviation", count(val) as "count" by field

[mfietz]

I'm impressed, that solved my issue!
Two small corrections: table prefix_* (Splunk changes hyphens to underscores when a field is recognized) and untable temp field val.
But apart from that: Brilliant! Thank you very much!

[mfietz]

fieldsummary seems to mostly do what I want, but it is missing percentiles and median