thanks for the tip! i would like to take this opportunity to point out that with metrics and time series, even your finest "granularity" has already discarded whole worlds of useful data and relationships that you can never get back. metrics are a slaughterhouse for context.
like i've had people get all sniffy about sampling, like "WE need EVERY request"
bullshit. you already drop all the connective tissue of your data by using metrics. so you'd rather retain 100% of atomized meaningless values, not keep a representative sample in usable form? 🙄
bullshit. you already drop all the connective tissue of your data by using metrics. so you'd rather retain 100% of atomized meaningless values, not keep a representative sample in usable form? 🙄
news flash: we all delete data. or discard or drop or ignore it.
We all. Delete. Data.
this needs to be a conversation about tradeoffs, no purists.
We all. Delete. Data.
this needs to be a conversation about tradeoffs, no purists.
if you're developing a transactional system or a replication log, then yes, please keep your data schema'd and compact and predictable. add friction around changes. replicate and backup and retain the living shit out of it.
if you're dealing with operational data, do the opposite of all those things.
be profligate. be wide, be dense, be lazy and lossy and accept generously. incentivize developers to toss more shit in ad hoc without having to run migration or change schemas. welcome all comers.
be profligate. be wide, be dense, be lazy and lossy and accept generously. incentivize developers to toss more shit in ad hoc without having to run migration or change schemas. welcome all comers.
... a ha! i DID already rant about this.
