[ 
https://issues.apache.org/jira/browse/RANGER-5875?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Dineshkumar Yadav updated RANGER-5875:
--------------------------------------
    Fix Version/s: 3.0.0

> OpenSearch access audit search misses accessType values that differ in case
> ---------------------------------------------------------------------------
>
>                 Key: RANGER-5875
>                 URL: https://issues.apache.org/jira/browse/RANGER-5875
>             Project: Ranger
>          Issue Type: Task
>          Components: Ranger
>            Reporter: Dineshkumar Yadav
>            Assignee: Dineshkumar Yadav
>            Priority: Major
>             Fix For: 3.0.0
>
>          Time Spent: 20m
>  Remaining Estimate: 0h
>
> {{GET /service/assets/accessAudit?accessType=...}} returns no rows from 
> OpenSearch when the stored access type does not match the lowercased, 
> Solr-escaped query. The same filter works against Solr.
> Hive and other plugins write the access type as produced by the authorizer, 
> for example {{SELECT}} or {{{}METADATA OPERATION{}}}. 
> {{OpenSearchAuditDestination}} stores that string on the {{access}} field. 
> The OpenSearch mapping defines {{access}} as a {{{}keyword{}}}, so the 
> indexed token keeps the original case. Solr defines the same field as 
> {{key_lower_case}} (keyword tokenizer plus lowercase filter), and 
> {{SolrUtil}} queries it as a Lucene query string, so {{accessType=SELECT}} 
> becomes {{access:select}} and matches.
> The OpenSearch query path lowercases the parameter, escapes it with Solr's 
> {{{}ClientUtils.escapeQueryChars{}}}, and sends a {{{}match_phrase{}}}. 
> {{SELECT}} is queried as {{{}select{}}}, which does not match the keyword 
> {{{}SELECT{}}}. {{METADATA OPERATION}} is queried as {{{}metadata\ 
> operation{}}}. A {{match_phrase}} treats the backslash as part of the text, 
> so that value does not match either. The database audit search is also 
> different: it uses a partial match on {{{}accessType{}}}.
> Fix:
>  * Build exact string filters, except the {{text}} field {{{}reason{}}}, as a 
> case-insensitive {{term}} query on the original value. Do not Solr-escape 
> that value.
>  * Add a {{key_lower_case}} normalizer (custom normalizer with the 
> {{lowercase}} filter) to the keyword audit fields that Solr types as 
> {{{}key_lower_case{}}}. Leave {{id}} as a plain {{{}keyword{}}}, matching 
> Solr's case-sensitive {{string}} field.
>  * Register that normalizer in index settings when 
> {{OpenSearchIndexBootStrapper}} creates the audit index. The schema file is 
> loaded as mappings only, so a mapping that names {{key_lower_case}} fails 
> unless the setting exists.
> The normalizer applies only to an index created after this change. The 
> {{term}} query is still required: a normalizer does not stop the old escaped 
> {{match_phrase}} from missing values that contain spaces or query characters.
> Test plan:
>  * Create a new OpenSearch audit index and confirm {{access}} uses the 
> {{key_lower_case}} normalizer.
>  * Write audits with access types {{SELECT}} and {{{}METADATA OPERATION{}}}.
>  * {{{}accessType=SELECT{}}}, {{{}accessType=select{}}}, and 
> {{accessType=METADATA OPERATION}} each return the matching rows.
>  * A {{reason}} filter still matches.
>  * Solr audit search behavior is unchanged.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to