Carbondata can provide some way to disable usage of multi-layered index and provide full table scan.
This may help in following cases; 1. Small tables occupying only few number of blocks are probably better of using full table scan. 2. Queries with large number of projections with no filter may benefit from using full table scan. 3. Testing different scenarios and comparing with multiple HDFS file formats that do not provide multi layer index will be easier. Also, Carbondata scan internally should be smart enough to detect this based on query, data size, etc. Any comments? |
Administrator
|
Hi
good suggestion! Add one configurable option to disable index for no filter and small table scenarios. One comment : you are suggesting only disable MDK; How about other index(inverted index,and MINMAX index)? +1 for this feature Regards Liang 2016-09-28 6:49 GMT+08:00 Aniket Adnaik <[hidden email]>: > Carbondata can provide some way to disable usage of multi-layered index and > provide full table scan. > This may help in following cases; > 1. Small tables occupying only few number of blocks are probably better of > using full table scan. > 2. Queries with large number of projections with no filter may benefit from > using full table scan. > 3. Testing different scenarios and comparing with multiple HDFS file > formats > that do not provide multi layer index will be easier. > > Also, Carbondata scan internally should be smart enough to detect this > based > on query, data size, etc. > > Any comments? > > > > > -- > View this message in context: http://apache-carbondata- > mailing-list-archive.1130556.n5.nabble.com/Discussion- > option-to-disable-multi-layered-index-scan-and-use- > full-table-scan-tp1526.html > Sent from the Apache CarbonData Mailing List archive mailing list archive > at Nabble.com. > -- Regards Liang |
Ideally this should be an internal improvement, not necessarily exposing it as an config option, Carbon should be able to smartly figure out if leveraging index is beneficial or straightly going for a file scan (just as Parquet).
Regards. Jihong -----Original Message----- From: Liang Big data [mailto:[hidden email]] Sent: Tuesday, September 27, 2016 4:43 PM To: [hidden email] Subject: Re: [Discussion]: option to disable multi-layered index scan and use full table scan Hi good suggestion! Add one configurable option to disable index for no filter and small table scenarios. One comment : you are suggesting only disable MDK; How about other index(inverted index,and MINMAX index)? +1 for this feature Regards Liang 2016-09-28 6:49 GMT+08:00 Aniket Adnaik <[hidden email]>: > Carbondata can provide some way to disable usage of multi-layered index and > provide full table scan. > This may help in following cases; > 1. Small tables occupying only few number of blocks are probably better of > using full table scan. > 2. Queries with large number of projections with no filter may benefit from > using full table scan. > 3. Testing different scenarios and comparing with multiple HDFS file > formats > that do not provide multi layer index will be easier. > > Also, Carbondata scan internally should be smart enough to detect this > based > on query, data size, etc. > > Any comments? > > > > > -- > View this message in context: http://apache-carbondata- > mailing-list-archive.1130556.n5.nabble.com/Discussion- > option-to-disable-multi-layered-index-scan-and-use- > full-table-scan-tp1526.html > Sent from the Apache CarbonData Mailing List archive mailing list archive > at Nabble.com. > -- Regards Liang |
I agree with jihong.carbon need to have smart logic to decide
On Wed, 28 Sep 2016 at 6:12 AM, Jihong Ma <[hidden email]> wrote: > Ideally this should be an internal improvement, not necessarily exposing > it as an config option, Carbon should be able to smartly figure out if > leveraging index is beneficial or straightly going for a file scan (just as > Parquet). > > Regards. > > Jihong > > -----Original Message----- > From: Liang Big data [mailto:[hidden email]] > Sent: Tuesday, September 27, 2016 4:43 PM > To: [hidden email] > Subject: Re: [Discussion]: option to disable multi-layered index scan and > use full table scan > > Hi > > good suggestion! > Add one configurable option to disable index for no filter and small table > scenarios. > One comment : you are suggesting only disable MDK; How about other > index(inverted index,and MINMAX index)? > > +1 for this feature > > Regards > Liang > > 2016-09-28 6:49 GMT+08:00 Aniket Adnaik <[hidden email]>: > > > Carbondata can provide some way to disable usage of multi-layered index > and > > provide full table scan. > > This may help in following cases; > > 1. Small tables occupying only few number of blocks are probably better > of > > using full table scan. > > 2. Queries with large number of projections with no filter may benefit > from > > using full table scan. > > 3. Testing different scenarios and comparing with multiple HDFS file > > formats > > that do not provide multi layer index will be easier. > > > > Also, Carbondata scan internally should be smart enough to detect this > > based > > on query, data size, etc. > > > > Any comments? > > > > > > > > > > -- > > View this message in context: http://apache-carbondata- > > mailing-list-archive.1130556.n5.nabble.com/Discussion- > > option-to-disable-multi-layered-index-scan-and-use- > > full-table-scan-tp1526.html > > Sent from the Apache CarbonData Mailing List archive mailing list archive > > at Nabble.com. > > > > > > -- > > Regards > Liang > |
In my view we can have both;
1. We should have internal smartness built-in, like databases do based on cost estimation .Carbon can take simpler approach based on data size or number of files to scan. 2. Provide some kind of hints OR switch to enable or disable - MDK, MinMax and inverted index, individually and together. This would help us in testing different scenarios or performance tuning. These are like options for users who know what they are doing or can be internal options if you we don't want to create more confusion. Best Regards, Aniket On Tue, Sep 27, 2016 at 11:58 PM, Raghunandan S < [hidden email]> wrote: > I agree with jihong.carbon need to have smart logic to decide > On Wed, 28 Sep 2016 at 6:12 AM, Jihong Ma <[hidden email]> wrote: > > > Ideally this should be an internal improvement, not necessarily exposing > > it as an config option, Carbon should be able to smartly figure out if > > leveraging index is beneficial or straightly going for a file scan (just > as > > Parquet). > > > > Regards. > > > > Jihong > > > > -----Original Message----- > > From: Liang Big data [mailto:[hidden email]] > > Sent: Tuesday, September 27, 2016 4:43 PM > > To: [hidden email] > > Subject: Re: [Discussion]: option to disable multi-layered index scan and > > use full table scan > > > > Hi > > > > good suggestion! > > Add one configurable option to disable index for no filter and small > table > > scenarios. > > One comment : you are suggesting only disable MDK; How about other > > index(inverted index,and MINMAX index)? > > > > +1 for this feature > > > > Regards > > Liang > > > > 2016-09-28 6:49 GMT+08:00 Aniket Adnaik <[hidden email]>: > > > > > Carbondata can provide some way to disable usage of multi-layered index > > and > > > provide full table scan. > > > This may help in following cases; > > > 1. Small tables occupying only few number of blocks are probably better > > of > > > using full table scan. > > > 2. Queries with large number of projections with no filter may benefit > > from > > > using full table scan. > > > 3. Testing different scenarios and comparing with multiple HDFS file > > > formats > > > that do not provide multi layer index will be easier. > > > > > > Also, Carbondata scan internally should be smart enough to detect this > > > based > > > on query, data size, etc. > > > > > > Any comments? > > > > > > > > > > > > > > > -- > > > View this message in context: http://apache-carbondata- > > > mailing-list-archive.1130556.n5.nabble.com/Discussion- > > > option-to-disable-multi-layered-index-scan-and-use- > > > full-table-scan-tp1526.html > > > Sent from the Apache CarbonData Mailing List archive mailing list > archive > > > at Nabble.com. > > > > > > > > > > > -- > > > > Regards > > Liang > > > |
The use case scenario for direct full table scan is rare and not dominating case, the internal built-in smartness should be very cautious about making this decision as this is considered to be in the worst case category. still didn't see a need for exposing it externally.
Regards. Jenny -----Original Message----- From: Aniket Adnaik [mailto:[hidden email]] Sent: Wednesday, September 28, 2016 10:50 AM To: [hidden email] Subject: Re: [Discussion]: option to disable multi-layered index scan and use full table scan In my view we can have both; 1. We should have internal smartness built-in, like databases do based on cost estimation .Carbon can take simpler approach based on data size or number of files to scan. 2. Provide some kind of hints OR switch to enable or disable - MDK, MinMax and inverted index, individually and together. This would help us in testing different scenarios or performance tuning. These are like options for users who know what they are doing or can be internal options if you we don't want to create more confusion. Best Regards, Aniket On Tue, Sep 27, 2016 at 11:58 PM, Raghunandan S < [hidden email]> wrote: > I agree with jihong.carbon need to have smart logic to decide > On Wed, 28 Sep 2016 at 6:12 AM, Jihong Ma <[hidden email]> wrote: > > > Ideally this should be an internal improvement, not necessarily exposing > > it as an config option, Carbon should be able to smartly figure out if > > leveraging index is beneficial or straightly going for a file scan (just > as > > Parquet). > > > > Regards. > > > > Jihong > > > > -----Original Message----- > > From: Liang Big data [mailto:[hidden email]] > > Sent: Tuesday, September 27, 2016 4:43 PM > > To: [hidden email] > > Subject: Re: [Discussion]: option to disable multi-layered index scan and > > use full table scan > > > > Hi > > > > good suggestion! > > Add one configurable option to disable index for no filter and small > table > > scenarios. > > One comment : you are suggesting only disable MDK; How about other > > index(inverted index,and MINMAX index)? > > > > +1 for this feature > > > > Regards > > Liang > > > > 2016-09-28 6:49 GMT+08:00 Aniket Adnaik <[hidden email]>: > > > > > Carbondata can provide some way to disable usage of multi-layered index > > and > > > provide full table scan. > > > This may help in following cases; > > > 1. Small tables occupying only few number of blocks are probably better > > of > > > using full table scan. > > > 2. Queries with large number of projections with no filter may benefit > > from > > > using full table scan. > > > 3. Testing different scenarios and comparing with multiple HDFS file > > > formats > > > that do not provide multi layer index will be easier. > > > > > > Also, Carbondata scan internally should be smart enough to detect this > > > based > > > on query, data size, etc. > > > > > > Any comments? > > > > > > > > > > > > > > > -- > > > View this message in context: http://apache-carbondata- > > > mailing-list-archive.1130556.n5.nabble.com/Discussion- > > > option-to-disable-multi-layered-index-scan-and-use- > > > full-table-scan-tp1526.html > > > Sent from the Apache CarbonData Mailing List archive mailing list > archive > > > at Nabble.com. > > > > > > > > > > > -- > > > > Regards > > Liang > > > |
Free forum by Nabble | Edit this page |