Ranjith, If the schema of the data changes, when using external tables, you can drop the table and re-create it on the same dataset taking care of the schema changes (hopefully, maintaining backwards compatibility).
I think you can still achieve that using alter table commands with managed tables; however, I find external tables just easier to manage, so I almost always end up making all my HDFS tables external. Mark Mark Grover, Business Intelligence Analyst OANDA Corporation www: oanda.com www: fxtrade.com ----- Original Message ----- From: "Ranjith" <ranjith.raghuna...@gmail.com> To: user@hive.apache.org Cc: user@hive.apache.org Sent: Sunday, May 13, 2012 8:54:35 PM Subject: Re: Managed vs external tables in hive Thanks Mark and Edward. This is good info to keep in mind. So is it fair to say that external tables offer flexibility, in that, you can have multiple schemas on the same data asset without data duplication. Is there anything else that an external table may offer versus a hive managed table or vice versa? Thanks, Ranjith On May 13, 2012, at 6:47 PM, Edward Capriolo <edlinuxg...@gmail.com> wrote: > I believe I walked through the entire process. > > You can ALTER TABLE a table and change it from external to managed. So > someone could always change the table to MANAGED do the index thing > and then change it back. Just be aware of the tables current status > before it is dropped. > > Edward > > On Sun, May 13, 2012 at 4:07 PM, Ranjith <ranjith.raghuna...@gmail.com> wrote: >> Edward, >> Did you confirm this through the explain plan or through the execution of >> the ddl alone. And have you tried buckets with external tables? >> >> Thanks, >> Ranjith >> >> On May 13, 2012, at 2:33 PM, Edward Capriolo <edlinuxg...@gmail.com> wrote: >> >> The original design docs say you can not build indexes on external tables >> but I tried it in 0.8.x and confirmed you can. >> >> On Sunday, May 13, 2012, Ranjith <ranjith.raghunat h...@gmail.com> wrote: >>> Indexes can be built on tables managed by hive. For external tables I do >>> not believe that to be true. Please feel to correct if I am wrong. >>> >>> Thanks, >>> Ranjith >>> On May 12, 2012, at 9:24 PM, Nanda Vijaydev <nanda.vijay...@gmail.com> >>> wrote: >>> >>> In hive, the raw data is in HDFS and there is a metadata layer that >>> defines the structure of the raw data. Table is usually a reference to >>> metadata, probably in a mySQL server and it contains a reference to the >>> location of the data in HDFS, type of delimiter or serde to use and so on. >>> 1. With hive managed tables, when you drop a table, both the metadata in >>> mysql and raw data on the cluster gets deleted. >>> 2. With external tables, when you drop a table, just the metadata gets >>> deleted and the raw data continues to exist on the cluster. >>> >>> On Thu, May 10, 2012 at 3:02 PM, David Kulp <dk...@fiksu.com> wrote: >>>> >>>> It's simpler than this. All files look the same -- and are often very >>>> simple delimited text -- whether managed or external. The only difference >>>> is that the files associated with a managed table are dropped when the >>>> table >>>> is dropped and files that are loaded into a managed table are moved into >>>> hive's private path. External tables never move or remove files. >>>> Performance is the same. >>>> >>>> On May 10, 2012, at 5:52 PM, kulkarni.swar...@gmail.com wrote: >>>> >>>>> I am pretty new to hive and was trying to clearly understand the >>>>> difference between a managed and an external table. >>>>> >>>>> As my current understanding stands, a managed table is a table whose >>>>> data is completely owned by hive whereas an external table is usually >>>>> created to have a hive frontend for the data managed in external systems.I >>>>> would suppose this would mean that a query on an external table goes out >>>>> to >>>>> fetch data from the given external table, deserialize according to the >>>>> given/suitable SerDe and then show the output of the query in hive format. >>>>> >>>>> So does this mean that cost of using external tables is much higher >>>>> than the native ones? Or is there some caching that comes into play that I >>>>> am not seeing right now. >>>>> >>>>> Thanks for the help. >>>>> >>>>> -- >>>>> Swarnim >>>> >>> >>>
