Accessing External Data in AsterixDB

Table of Contents

• Introduction
• Adapter for an External Dataset
• Builtin Adapters
• Creating an External Dataset
• Writing Queries against an External Dataset
• Building Indexes over External Datasets
• External Data Snapshots
• Frequently Asked Questions

Introduction [Back to TOC]

Data that needs to be processed by AsterixDB could be residing outside AsterixDB storage. Examples include data files on a distributed file system such as HDFS or on the local file system of a machine that is part of an AsterixDB cluster. For AsterixDB to process such data, an end-user may create a regular dataset in AsterixDB (a.k.a. an internal dataset) and load the dataset with the data. AsterixDB also supports ‘‘external datasets’’ so that it is not necessary to “load” all data prior to using it. This also avoids creating multiple copies of data and the need to keep the copies in sync.

    create dataverse ExternalFileDemo;
    use dataverse ExternalFileDemo;

    create type LineitemType as closed {
      l_orderkey:int32,
      l_partkey: int32,
      l_suppkey: int32,
      l_linenumber: int32,
      l_quantity: double,
      l_extendedprice: double,
      l_discount: double,
      l_tax: double,
      l_returnflag: string,
      l_linestatus: string,
      l_shipdate: string,
      l_commitdate: string,
      l_receiptdate: string,
      l_shipinstruct: string,
      l_shipmode: string,
      l_comment: string}

    create external dataset Lineitem(LineitemType)
    using localfs

    use dataverse ExternalFileDemo;

    create external dataset Lineitem(LineitemType)
    using localfs
    (("path"="127.0.0.1://SOURCE_PATH"),
    ("format"="delimited-text"),
    ("delimiter"="|"));

    org.apache.hyracks.algebricks.common.exceptions.AlgebricksException: org.apache.hyracks.api.exceptions.HyracksDataException: org.apache.hyracks.api.exceptions.HyracksDataException: Job failed.

    127.0.0.1:///home/joe/lineitem.tbl.

    create external dataset Lineitem(LineitemType)
    using hdfs((“hdfs”:”hdfs://localhost:54310”),(“path”:”/asterix/Lineitem.tbl”),...,(“input- format”:”rc-format”));

    create external dataset Lineitem('LineitemType)
    using hdfs(("hdfs"="hdfs://localhost:54310"),("path"="/asterix/Lineitem.tbl"),("input-format"="text- input-format"),("format"="delimited-text"),("delimiter"="|"));

    create external dataset Lineitem('LineitemType)
    using hdfs(("hdfs"="hdfs://localhost:54310"),("path"="/asterix/SequenceLineitem.tbl"),("input- format"="sequence-input-format"),("format"="adm"));

    create external dataset Lineitem('LineitemType)
    using hdfs(("hdfs"="hdfs://localhost:54310"),("path"="/asterix/RCLineitem.tbl"),("input-format"="rc-input-format"),("format"="binary"),("parser"="hive-parser"),("hive- serde"="org.apache.hadoop.hive.serde2.columnar.ColumnarSerde"));

Writing Queries against an External Dataset [Back to TOC]

You may write AQL queries against an external dataset in exactly the same way that queries are written against internal datasets. The following is an example of an AQL query that applies a filter and returns an ordered result.

    use dataverse ExternalFileDemo;

    for $c in dataset('Lineitem')
    where $c.l_orderkey <= 3
    order by $c.l_orderkey, $c.l_linenumber
    return $c

Building Indexes over External Datasets [Back to TOC]

AsterixDB supports building B-Tree and R-Tree indexes over static data stored in the Hadoop Distributed File System. To create an index, first create an external dataset over the data as follows

    create external dataset Lineitem(LineitemType)
    using hdfs(("hdfs"="hdfs://localhost:54310"),("path"="/asterix/Lineitem.tbl"),("input-format"="text-input- format"),("format"="delimited-text"),("delimiter"="|"));

You can then create a B-Tree index on this dataset instance as if the dataset was internally stored as follows:

    create index PartkeyIdx on Lineitem(l_partkey);

You could also create an R-Tree index as follows:

    create index IndexName on DatasetName(attribute-name) type rtree;

After building the indexes, the AsterixDB query compiler can use them to access the dataset and answer queries in a more cost effective manner. AsterixDB can read all HDFS input formats, but indexes over external datasets can currently be built only for HDFS datasets with ‘text-input-format’, ‘sequence-input-format’ or ‘rc-input-format’.

External Data Snapshots [Back to TOC]

An external data snapshot represents the status of a dataset’s files in HDFS at a point in time. Upon creating the first index over an external dataset, AsterixDB captures and stores a snapshot of the dataset in HDFS. Only objects present at the snapshot capture time are indexed, and any additional indexes created afterwards will only contain data that was present at the snapshot capture time thus preserving consistency across all indexes of a dataset. To update all indexes of an external dataset and advance the snapshot time to be the present time, a user can use the refresh external dataset command as follows:

    refresh external dataset DatasetName;

After a refresh operation commits, all of the dataset’s indexes will reflect the status of the data as of the new snapshot capture time.

Frequently Asked Questions [Back to TOC]

Q. I added data to my dataset in HDFS, Will the dataset indexes in AsterixDB be updated automatically?

A. No, you must use the refresh external dataset statement to make the indexes aware of any changes in the dataset files in HDFS.

Q. Why doesn’t AsterixDB update external indexes automatically?

A. Since external data is managed by other users/systems with mechanisms that are system dependent, AsterixDB has no way of knowing exactly when data is added or deleted in HDFS, so the responsibility of refreshing indexes are left to the user. A user can use internal datasets for which AsterixDB manages the data and its indexes.

Q. I created an index over an external dataset and then added some data to my HDFS dataset. Will a query that uses the index return different results from a query that doesn’t use the index?

A. No, queries’ results are access path independent and the stored snapshot is used to determines which data are going to be included when processing queries.

Q. I created an index over an external dataset and then deleted some of my dataset’s files in HDFS, Will indexed data access still return the objects in deleted files?

A. No. When AsterixDB accesses external data, with or without the use of indexes, it only access files present in the file system at runtime.

Q. I submitted a refresh command on a an external dataset and a failure occurred, What has happened to my indexes?

A. External Indexes Refreshes are treated as a single transaction. In case of a failure, a rollback occurs and indexes are restored to their previous state. An error message with the cause of failure is returned to the user.

Q. I was trying to refresh an external dataset while some queries were accessing the data using index access method. Will the queries be affected by the refresh operation?

A. Queries have access to external dataset indexes state at the time where the queries are submitted. A query that was submitted before a refresh commits will only access data under the snapshot taken before the refresh; queries that are submitted after the refresh commits will access data under the snapshot taken after the refresh.

Q. What happens when I try to create an additional index while a refresh operation is in progress or vice versa?

A. The create index operation will wait until the refresh commits or aborts and then the index will be built according to the external data snapshot at the end of the refresh operation. Creating indexes and refreshing datasets are mutually exclusive operations and will not be run in parallel. Multiple indexes can be created in parallel, but not multiple refresh operations.