What were you using OpenTSDB for? Bigtable now supports a lot of time series capabilities out of the box e.g. there are
distributed counters for fast write time aggregations (sum/count, min/max, approximate count distinct) which you can also do for tumbling windows using date truncation on timestamps
continuous materialized views allow you to define more complex, multi-row aggregation logic using SQL that get automatically and incrementally updated as new writes come. You can also re-key your data using these views for secondary access patterns like building a secondary index
As part of Bigtable's SQL support you can also do read-time aggregations on top of the pre-aggregated data e.g. if you preaggregate to hourly using counters or incremental materialized views, you can filter/group to daily, weekly etc. at read time using a GROUP BY at read time including merging data sketches e.g. from daily active users to monthly active users etc.
These are most common operations in time series databases and doing directly in the database would simplify your stack instead of running an additional service on top of it in GKE.