Dear Wiki user, You have subscribed to a wiki page or wiki category on "Solr Wiki" for change notification.
The following page has been changed by Lance Norskog: http://wiki.apache.org/solr/UniqueKey ------------------------------------------------------------------------------ The Solr '''uniqueKey''' field encodes the ''identity semantics'' of a document. In database jargon, the ''primary key''. - = Use cases = + A Solr index does not need a unique key, but almost all indexes use one. + == Use cases which do not require a unique key == - You do not necessarily need a unique key to make an index, but almost all indexes use one. - - '''In these use cases, no unique key is necessary.''' * Build an index from empty. Search for documents. When you have new documents to add, either add them or clear index and reindex from scratch. You know that you will never add the same document twice. * Sort documents found in a search against a numerical field. + == Use cases which require a unique key == - - '''These use cases require a unique key.''' * Add documents incrementally. You do not rebuild the index from scratch but want to add new documents periodically. You may add the same document twice and it will only be stored once. An example is an RSS feed from a blog: the RSS feed will be polled, therefore the same article may appear more than once. * Do statistical analysis on indexed data. @@ -21, +18 @@ There is a saying in database design:''data sticks where it lands''. Once you store data in some format and container, it is very hard to change this decision. By adding a layer of indirection in the SOLR schema's identity, you give yourself the ability to change the innate identity of the document. * Multiple queries about the same document, with document id saved for future reference. * Delete documents. + == Use cases which require a unique key generated from data in the document == - - '''For these use cases you need to generate the key from fields in the document. * Allow different database systems to create identity keys that work in other systems. The documents may come from multiple sources, and be stored in multiple places. There may not be one convenient place in the indexing path to create a unique id. The different sources will need to separately implement the same algorithm. The key should be a short unique string (see UUID below). - = Unique key composition = There are different data sources available for the unique key: - * Raw text native to the document + * Text in the document + * UUID key generated from the time of insertion and a random number * UUID data generated from data in the document + == Text field in the document == + In the blog RSS example above, the URL of each article. The field must be single-valued. + == UUID techniques == - * UUID key generated from the time of insertion and a random number - - Only one field in the document may be used as the unique key. It cannot be multivalued. - - == UUID == - UUID is short for Universal Unique IDentifier. The UUID standard [http://www.ietf.org/rfc/rfc4122.txt RFC-4122] includes several types of UUID with different input formats. There is a UUID field type in Solr 1.4 which implements version 4. Also, the ExtractingRequestHandler automatically creates UUID version 4. + UUID is short for Universal Unique IDentifier. The UUID standard [http://www.ietf.org/rfc/rfc4122.txt RFC-4122] includes several types of UUID with different input formats. There is a UUID field type in Solr 1.4 which implements version 4. Also, the ExtractingRequestHandler automatically creates UUID version 4. You can also implement a UUID string from a cryptographic hash. - == Cryptographic hash == - A cryptographic hashing algorithm can be thought of as creating N very random bits from the input data. The MD5 algorithm create 128 bits. This means that 2 input data sets have a chance of 1 in 2^128 of creating the same MD5. There is a standard expression of this as 32 hexadecimal characters. [http://www.ietf.org/rfc/rfc1321.txt RFC-1321]. Several MD5 digest algorithm packages for various languages do not follow this standard. The UUID standard always includes the time at the creation of the UUID, which precludes some of the above use cases. You can cheat and ignore the clock requirement. It is probably better to use the UUID type than a raw string. + A cryptographic hashing algorithm can be thought of as creating N very random bits from the input data. The MD5 algorithm create 128 bits. This means that 2 input data sets have a chance of 1 in 2^128 of creating the same MD5. There is a standard expression of this as 32 hexadecimal characters. [http://www.ietf.org/rfc/rfc1321.txt RFC-1321]. Several MD5 digest algorithm packages for various languages do not follow this standard. The UUID standard always includes the time at the creation of the UUID, which precludes some of the above use cases. You can cheat and ignore the clock requirement. It is best to use the UUID text format: ''550e8400-e29b-41d4-a716-446655440000'' instead of ''550e8400e29b41d4a716446655440000''. (You will read many of these keys.)[[BR]] - One advantage in using a crypto-generated unique key is that you can select a random subset of documents via wildcards. If the UUID data is saved as a string in the 32-character RFC format, 'd3adbe3fdeadb3e4deadbee4deadb3ef', the query "id:a*" will select a random 1/16 of the entire document set. "id:aa*" selects 1/256 of the document set, again very randomly. Statistical analysis and data extraction projects can use this to select small subsets instead of walking the entire index.
