Thank you for making this proposal Yung-An He. What does "...semantic analysis using the tree data structure..." mean?
The hbase:tableschema would specify column typing? It would allow compound types? How about the row key as compound type? What typing system will HBaseQL use? Will it make a new one or use Phoenix types or the hbase OrderedType? Having to describe the table in hbase:tableschema is similar to how Drill operates; we could embed a Drillbit and get what is here in this proposal? I wouldn't lean too much on the need of a restart differentiating HBaseQL and Phoenix or even the fact that Phoenix owns its tables; in the scheme of things these inconveniences fade after initial one-time pain. Upsides as I see it are table-scope and no joins; i.e. basic. I see some downside though + Yet another HBase SQL (YAHS) in an already crowded space offering a new permutation that is intentionally crimped and so likely useful at first but ultimately may just frustrate (no joins, etc.) + It is a new effort rather when existing projects are starved for resources as it is (e.g. could this HBaseQL be done using a cut-down, table-scoped, no-joins Phoenix so Phoenix got the benefit of some dev resource? Or as an embedded Drillbit with HBase amenities?) + The user would have to run two shells: hbase sql and hbase shell. + It is a load of new code in a new domain (SQL) that we'd be pulling into a hbase, a project that is already broad in scope. Thanks again for putting up the proposal Yung-An He, St.Ack On Wed, Sep 27, 2017 at 7:27 AM, Yung-An He <mathst...@gmail.com> wrote: > Hi folks, > > Currently, HBase hasn’t support SQL syntax yet. Many users who are > familiar with SQL syntax can only query data from HBase via Hive, Impala, > Phoenix or other tools that support SQL syntax. However, some tools are too > complicated to install and need to restart the HBase cluster. Additionally, > there may be side effects that affect raw data. If HBase has a native SQL > querying module which is easy to install without restarting the cluster as > well as not affecting the original data, it would be extremely helpful and > handy for users that are already familiar with SQL syntax. > > In HareDB, we have implemented HareQL, and hope that we can contribute > some piece as a module to HBase. See attached document for more details. >
