tsd.query.skip_unresolved_tagvs causes performance issues?

[Huafeng Lyu]

Hi,

We've just enabled tsd.query.skip_unresolved_tagvs to avoid "no such name for tagv ..." exception when querying with tagvs that still doesn't exist yet, but from our observation, this seems caused much more hbase read requests and thus much longer query response time.

Can experts explain how does skip_unresolved_tagvs work? What's the best practice to avoid this exception while not affecting the performance? The workaround we're using now is to try to create these tagvs by using api/uid/assign before any queries are issued, but not sure if this is a good idea.

Thanks!

--huafeng

[ManOLamancha]

Ah ok so what happens here is that we don't have a negative cache (something we should add) in that each time a request contains a tag that hasn't been assigned, it will always call to HBase for the data. For the tags that DO exist, they're in the TSD's local cache.

If you don't mind filing a Github issue we can look to add that.