Access History¶
Access History requires Enterprise Edition (or higher). To inquire about upgrading, please contact Snowflake Support.
Preview Feature — Open
Access History is generally available to all accounts that are Enterprise Edition (or higher).
The POLICIES_REFERENCED column is in preview.
This topic provides concepts on the user access history in Snowflake.
Overview¶
Access History in Snowflake refers to when the user query reads column data and when the SQL statement performs a data write operation, such as INSERT, UPDATE, and DELETE along with variations of the COPY command, from the source data object to the target data object. The user access history can be found by querying the Account Usage ACCESS_HISTORY view.
Each row in the ACCESS_HISTORY view contains a single record per SQL statement. The record describes the columns the query accessed directly and indirectly (i.e. the underlying tables that the data for the query comes from). These records facilitate regulatory compliance auditing and provide insights on popular and frequently accessed tables and columns since there is a direct link between the user (i.e. query operator), the query, the table or view, the column, and the data.
Note
Records in the Account Usage QUERY_HISTORY view do not always get recorded in the ACCESS_HISTORY view. The structure of the SQL statement determines whether Snowflake records an entry in the ACCESS_HISTORY view.
For details on the read and write operations Snowflake supports in the ACCESS_HISTORY view, refer to the view Usage Notes.
Tracking Read and Write Operations¶
The ACCESS_HISTORY view contains these columns:
QUERY_ID | QUERY_START_TIME | USER_NAME | DIRECT_OBJECTS_ACCESSED | BASE_OBJECTS_ACCESSED | OBJECTS_MODIFIED | OBJECT_MODIFIED_BY_DDL | POLICIES_REFERENCED
Read operations are tracked through the first five columns, while the last column, OBJECTS_MODIFIED, specifies the data write information that involved Snowflake columns, tables, and stages.
The query in Snowflake and how the database objects were created determines the information Snowflake returns in the DIRECT_OBJECTS_ACCESSED, BASE_OBJECTS_ACCESSED, and OBJECTS_MODIFIED columns.
Similarly, if the query references an object protected by a row access policy or a column protected by a masking policy, Snowflake records the policy information in the POLICIES_REFERENCED column.
For column details, see the Columns section in the ACCESS_HISTORY view.
Read¶
Consider the following scenario to understand a read query and how the ACCESS_HISTORY view records this information:
A series of objects:
base_table»view_1»view_2»view_3.A read query on
view_2, such as:select * from view_2;
In this example, Snowflake returns:
view_2in the DIRECT_OBJECTS_ACCESSED column because the query specifiesview_2.base_tablein the BASE_OBJECTS_ACCESSED column because that is the original source of the data inview_2.
Note that view_1 and view_3 are not included in the DIRECT_OBJECTS_ACCESSED and BASE_OBJECTS_ACCESSED columns because neither of
those views were included in the query and they are not the base object that serves as the source for the data in view_2.
Write¶
Consider the following scenario to understand a write operation and how the ACCESS_HISTORY view records this information:
A data source:
base_tableCreate a table from the data source (i.e. CTAS):
create table table_1 as select * from base_table;
In this example, Snowflake returns:
base_tablein the BASE_OBJECTS_ACCESSED and DIRECT_OBJECTS_ACCESSED columns because the table was accessed directly and is the source of the data.table_1in the OBJECTS_MODIFIED column with the columns that were written to when creating the table.
Column Lineage¶
Column lineage (i.e. access history for columns) extends the Account Usage ACCESS_HISTORY view to specify how data flows from the source column to the target column in a write operation. Snowflake tracks the data from the source columns through all subsequent table objects that reference data from the source columns (e.g. INSERT, MERGE, CTAS) provided that objects in the lineage chain are not dropped. Snowflake makes column lineage accessible by enhancing the OBJECTS_MODIFIED column in the ACCESS_HISTORY view.
For additional details, see:
Benefits (in this topic)
Example: Column Lineage (in this topic)
Masking & Row Access Policy References¶
The POLICY_REFERENCED column specifies the object that has a row access policy set on a table or a masking policy set on a column, including any intermediate objects that are protected by either a row access policy or a masking policy. Snowflake records the policy that is enforced on the table or column.
Consider these objects:
t1 » v1 » v2
Where:
t1is a base table.v1is a view built from the base table.v2is a view built fromv1.
If the user queries v2, the POLICIES_REFERENCED column records either the row access policy that protects v2, each masking policy
that protects the columns in v2, or both kinds of policy as applicable. Additionally, this column records any masking or row access
policies that protect t1 and v1.
These records can help data governors understand how their policy-protected objects are accessed.
The POLICIES_REFERENCED column provides additional benefits to the ACCESS_HISTORY view:
Identify the policy-protected objects a user accesses in a given query.
Simplify the policy audit process.
Querying the ACCESS_HISTORY view eliminates the need for complex joins on other Account Usage views (e.g. POLICY_REFERENCES and QUERY_HISTORY), to obtain information about the protected objects and protected columns a user accesses.
Supported Operations¶
For a complete description of the read and write operations the ACCESS_HISTORY view supports, see the usage notes sections in the ACCESS_HISTORY View.
Benefits¶
Access history in Snowflake provides the following benefits pertaining to read and write operations:
- Data discovery
Discover unused data to determine whether to archive or delete the data.
- Track how sensitive data moves
Track data movement from an external cloud storage location (e.g. Amazon S3 bucket) to the target Snowflake table, and vice versa.
Track internal data movement from a Snowflake table to a different Snowflake table.
After tracing the movement of sensitive data, apply policies (masking and row access) to protect data, update access control settings to further regulate access to the stage and table, and set tags to ensure stages, tables, and columns with sensitive data can be tracked for compliance requirements.
- Data validation
The accuracy and integrity of reports, dashboards, and data visualization products such as charts and graphs are validated since the data can be traced to its original source.
Data stewards can also notify users prior to dropping or altering a given table or view.
- Compliance auditing
Identify the Snowflake user who performed a write operation on a table or stage and when the write operation occurred to meet compliance regulations, such as GDPR and CCPA.
- Enhance overall data governance
The ACCESS_HISTORY view provides a unified picture of what data was accessed, when the data access took place, and how the accessed data moved from the data source object to the data target object.
Column lineage provides these additional benefits:
- Protect Derived Objects
Data stewards can easily tag sensitive source columns without having to do additional work after creating derived objects (e.g. CTAS). Subsequently, the data steward can protect tables containing sensitive columns with a row access policy or protect the sensitive columns themselves with either a masking policy or a tag-based masking policy.
- Sensitive Column Copy Frequency
Data privacy officers can quickly determine the object count (e.g. 1 table, 2 views) of a column containing sensitive data. By knowing how many times a column with sensitive data appears in a table object, data privacy officers can prove how they satisfy regulatory compliance standards (e.g. to meet General Data Protection Regulation (GDPR) standards in the European Union).
- Root Cause Analysis
Column lineage provides a mechanism to trace the data to its source, which can help to pinpoint points of failure resulting from poor data quality and reduce the number of columns to analyze during the troubleshooting process.
Querying the ACCESS_HISTORY View¶
The following sections provide example queries for the ACCESS_HISTORY view.
Note that the some of the example queries filter on the QUERY_START_TIME column to increase query performance. Another option to increase performance is to query over narrower time ranges.
Examples: Read Queries¶
The subsections below detail how to query the ACCESS_HISTORY view for read operations for the following use cases:
Obtain the access history for a specific user.
Facilitate compliance audits for sensitive data access in the last 30 days, based on
object_id(e.g. a table id), to answer the following questions:Who accessed the data?
When was the data accessed?
What columns were accessed?
Return the User Access History¶
Return the user access history, ordered by user and query start time, starting from the most recent access.
SELECT user_name , query_id , query_start_time , direct_objects_accessed , base_objects_accessed FROM access_history ORDER BY 1, 3 desc ;
Facilitate Compliance Audits¶
The following examples help to facilitate compliance audits:
Add the
object_idvalue to determine who accessed a sensitive table in the last 30 days:SELECT distinct user_name FROM access_history , lateral flatten(base_objects_accessed) f1 WHERE f1.value:"objectId"::int=<fill_in_object_id> AND f1.value:"objectDomain"::string='Table' AND query_start_time >= dateadd('day', -30, current_timestamp()) ;
Using the
object_idvalue of32998411400350, determine when the access occurred in the last 30 days:SELECT query_id , query_start_time FROM access_history , lateral flatten(base_objects_accessed) f1 WHERE f1.value:"objectId"::int=32998411400350 AND f1.value:"objectDomain"::string='Table' AND query_start_time >= dateadd('day', -30, current_timestamp()) ;
Using the
object_idvalue of32998411400350, determine which columns were accessed in the last 30 days:SELECT distinct f4.value AS column_name FROM access_history , lateral flatten(base_objects_accessed) f1 , lateral flatten(f1.value) f2 , lateral flatten(f2.value) f3 , lateral flatten(f3.value) f4 WHERE f1.value:"objectId"::int=32998411400350 AND f1.value:"objectDomain"::string='Table' AND f4.key='columnName' ;
Examples: Write Operations¶
The subsections below detail how to query the ACCESS_HISTORY view for write operations for the following use cases:
Load data from a stage to a table.
Unload data from a table to a stage.
Use the PUT command to upload a local file to a stage.
Use the GET command to retrieve data files from a stage to a local directory.
Tracking sensitive stage data movement.
Load Data from a Stage to a Table¶
Load a set of values from a data file in external cloud storage into columns in a target table.
copy into table1(col1, col2) from (select t.$1, t.$2 from @mystage1/data1.csv.gz);
The DIRECT_OBJECTS_ACCESSED and BASE_OBJECTS_ACCESSED column specify that an external named stage was accessed:
{ "objectDomain": STAGE "objectName": "mystage1", "objectId": 1, "stageKind": "External Named" }
The OBJECTS_MODIFIED column specifies that data was written to two columns of the table:
{ "columns": [ { "columnName": "col1", "columnId": 1 }, { "columnName": "col2", "columnId": 2 } ], "objectId": 1, "objectName": "TEST_DB.TEST_SCHEMA.TABLE1", "objectDomain": TABLE }
Unload Data from a Table to a Stage¶
Unload a set of values from a Snowflake table into cloud storage.
copy into @mystage1/data1.csv from table1;
The DIRECT_OBJECTS_ACCESSED and BASE_OBJECTS_ACCESSED columns specify the table columns that were accessed:
{ "objectDomain": TABLE "objectName": "TEST_DB.TEST_SCHEMA.TABLE1", "objectId": 123, "columns": [ { "columnName": "col1", "columnId": 1 }, { "columnName": "col2", "columnId": 2 } ] }
The OBJECTS_MODIFIED column specifies the stage to which the accessed data was written:
{ "objectId": 1, "objectName": "mystage1", "objectDomain": STAGE, "stageKind": "External Named" }
Use the PUT Command to Upload a Local File to a Stage¶
Copy a data file to an internal (i.e. Snowflake) stage.
put file:///tmp/data/mydata.csv @my_int_stage;
The DIRECT_OBJECTS_ACCESSED and BASE_OBJECTS_ACCESSED columns specify the local path to the file that was accessed:
{ "location": "file:///tmp/data/mydata.csv" }
The OBJECTS_MODIFIED column specifies the stage where the accessed data was written:
{ "objectId": 1, "objectName": "my_int_stage", "objectDomain": STAGE, "stageKind": "Internal Named" }
Use the GET Command to Retrieve Data Files from a Stage to a Local Directory¶
Retrieve a data file from an internal stage to a directory on the local machine.
get @%mytable file:///tmp/data/;
The DIRECT_OBJECTS_ACCESSED and BASE_OBJECTS_ACCESSED columns specify the stage and local directory that were accessed:
{ "objectDomain": Stage "objectName": "mytable", "objectId": 1, "stageKind": "Table" }
The OBJECTS_MODIFIED column specifies the directory to which the accessed data was written:
{ "location": "file:///tmp/data/" }
Tracking Sensitive Stage Data Movement¶
Track sensitive stage data as it moves through a series of queries executed in chronological order.
Execute the following queries. Note that five of the statements access stage data. Therefore, when you query the ACCESS_HISTORY view for stage access, the result set should include five rows.
use test_db.test_schema; create or replace table T1(content variant); insert into T1(content) select parse_json('{"name": "A", "id":1}'); -- T1 -> T6 insert into T6 select * from T1; -- S1 -> T1 copy into T1 from @S1; -- T1 -> T2 create table T2 as select content:"name" as name, content:"id" as id from T1; -- T1 -> S2 copy into @S2 from T1; -- S1 -> T3 create or replace table T3(customer_info variant); copy into T3 from @S1; -- T1 -> T4 create or replace table T4(name string, id string, address string); insert into T4(name, id) select content:"name", content:"id" from T1; -- T6 -> T7 create table T7 as select * from T6;Where:
T1,T2…T7specify the names of tables.
S1andS2specify the names of stages.
Query the access history to determine the access to stage S1.
The data for the DIRECT_OBJECTS_ACCESSED, BASE_OBJECTS_ACCESSED, and OBJECTS_MODIFIED columns are shown in the following table.
DIRECT_OBJECTS_ACCESSED
BASE_OBJECTS_ACCESSED
OBJECTS_MODIFIED
Note the following about the query example:
Uses a JOIN construct rather than a USING clause.
with access_history_flatten as ( select r.value:"objectId" as source_id, r.value:"objectName" as source_name, r.value:"objectDomain" as source_domain, w.value:"objectId" as target_id, w.value:"objectName" as target_name, w.value:"objectDomain" as target_domain, c.value:"columnName" as target_column, t.query_start_time as query_start_time from (select * from TEST_DB.ACCOUNT_USAGE.ACCESS_HISTORY) t, lateral flatten(input => t.BASE_OBJECTS_ACCESSED) r, lateral flatten(input => t.OBJECTS_MODIFIED) w, lateral flatten(input => w.value:"columns", outer => true) c ), sensitive_data_movements(path, target_id, target_name, target_domain, target_column, query_start_time) as -- Common Table Expression ( -- Anchor Clause: Get the objects that access S1 directly select f.source_name || '-->' || f.target_name as path, f.target_id, f.target_name, f.target_domain, f.target_column, f.query_start_time from access_history_flatten f where f.source_domain = 'Stage' and f.source_name = 'TEST_DB.TEST_SCHEMA.S1' and f.query_start_time >= dateadd(day, -30, date_trunc(day, current_date)) union all -- Recursive Clause: Recursively get all the objects that access S1 indirectly select sensitive_data_movements.path || '-->' || f.target_name as path, f.target_id, f.target_name, f.target_domain, f.target_column, f.query_start_time from access_history_flatten f join sensitive_data_movements on f.source_id = sensitive_data_movements.target_id and f.source_domain = sensitive_data_movements.target_domain and f.query_start_time >= sensitive_data_movements.query_start_time ) select path, target_name, target_id, target_domain, array_agg(distinct target_column) as target_columns from sensitive_data_movements group by path, target_id, target_name, target_domain;The query produces the following result set related to stage
S1data movement:
PATH
TARGET_NAME
TARGET_ID
TARGET_DOMAIN
TARGET_COLUMNS
TEST_DB.TEST_SCHEMA.S1–>TEST_DB.TEST_SCHEMA.T1
TEST_DB.TEST_SCHEMA.T1
66564
Table
[“CONTENT”]
TEST_DB.TEST_SCHEMA.S1–>TEST_DB.TEST_SCHEMA.T1–>TEST_DB.TEST_SCHEMA.S2
TEST_DB.TEST_SCHEMA.S2
118
Stage
[]
TEST_DB.TEST_SCHEMA.S1–>TEST_DB.TEST_SCHEMA.T1–>TEST_DB.TEST_SCHEMA.T2
TEST_DB.TEST_SCHEMA.T2
66568
Table
[“NAME”,”ID”]
TEST_DB.TEST_SCHEMA.S1–>TEST_DB.TEST_SCHEMA.T1–>TEST_DB.TEST_SCHEMA.T4
TEST_DB.TEST_SCHEMA.T4
66572
Table
[“ID”,”NAME”]
TEST_DB.TEST_SCHEMA.S1–>TEST_DB.TEST_SCHEMA.T3
TEST_DB.TEST_SCHEMA.T3
66570
Table
[“CUSTOMER_INFO”]
Example: Column Lineage¶
The following example queries the ACCESS_HISTORY view and uses the FLATTEN function to flatten the OBJECTS_MODIFIED column.
As a representative example, execute the following SQL query in your Snowflake account to produce the table below, where the numbered comments indicate the following:
// 1: Get the mapping between thedirectSourcesfield and the target column.// 2: Get the mapping between thebaseSourcesfield and the target column.
// 1
select
directSources.value: "objectId" as source_object_id,
directSources.value: "objectName" as source_object_name,
directSources.value: "columnName" as source_column_name,
'DIRECT' as source_column_type,
om.value: "objectName" as target_object_name,
columns_modified.value: "columnName" as target_column_name
from
(
select
*
from
snowflake.account_usage.access_history
) t,
lateral flatten(input => t.OBJECTS_MODIFIED) om,
lateral flatten(input => om.value: "columns", outer => true) columns_modified,
lateral flatten(
input => columns_modified.value: "directSources",
outer => true
) directSources
union
// 2
select
baseSources.value: "objectId" as source_object_id,
baseSources.value: "objectName" as source_object_name,
baseSources.value: "columnName" as source_column_name,
'BASE' as source_column_type,
om.value: "objectName" as target_object_name,
columns_modified.value: "columnName" as target_column_name
from
(
select
*
from
snowflake.account_usage.access_history
) t,
lateral flatten(input => t.OBJECTS_MODIFIED) om,
lateral flatten(input => om.value: "columns", outer => true) columns_modified,
lateral flatten(
input => columns_modified.value: "baseSources",
outer => true
) baseSources
;
Returns:
SOURCE_OBJECT_ID
SOURCE_OBJECT_NAME
SOURCE_COLUMN_NAME
SOURCE_COLUMN_TYPE
TARGET_OBJECT_NAME
TARGET_COLUMN_NAME
1
D.S.T0
NAME
BASE
D.S.T1
NAME
2
D.S.V1
NAME
DIRECT
D.S.T1
NAME
Example: Track Row Access Policy References¶
Return a row for each instance when a row access policy is set on a table, view, or materialized view without duplicates:
use role accountadmin; select distinct obj_policy.value:"policyName"::VARCHAR as policy_name from snowflake.account_usage.access_history as ah , lateral flatten(ah.policies_referenced) as obj , lateral flatten(obj.value:"policies") as obj_policy ;
Example: Track Masking Policy References¶
Return a row for each instance when a masking policy protects a column without duplicates. Note that additional flattening is necessary because the POLICIES_REFERENCED columns specifies the masking policy on a column one level deeper than the row access policy on a table:
use role accountadmin; select distinct policies.value:"policyName"::VARCHAR as policy_name from snowflake.account_usage.access_history as ah , lateral flatten(ah.policies_referenced) as obj , lateral flatten(obj.value:"columns") as columns , lateral flatten(columns.value:"policies") as policies ;
Examples: UDFs¶
These UDF examples show how the Account Usage ACCESS_HISTORY view records:
Calling a UDF named
get_product.Insert the product of calling the
get_productfunction into a table namedmydb.tables.t1.Shared UDFs.
Call a UDF¶
Consider the following SQL UDF that calculates the product of two numbers and assume it is stored in the schema named mydb.udfs:
CREATE FUNCTION MYDB.UDFS.GET_PRODUCT(num1 number, num2 number) RETURNS number AS $$ NUM1 * NUM2 $$ ;
Calling get_product directly results in recording the UDF details in the DIRECT_OBJECTS_ACCESSED
column:
[ { "objectDomain": "FUNCTION", "objectName": "MYDB.UDFS.GET_PRODUCT", "objectId": "2", "argumentSignature": "(NUM1 NUMBER, NUM2 NUMBER)", "dataType": "NUMBER(38,0)" } ]
This example is analogous to calling a stored procedure (in this topic).
UDF with INSERT DML¶
Consider the following INSERT statement to update the columns named 1 and 2 in the table named mydb.tables.t1:
insert into t1(product) select get_product(c1, c2) from mydb.tables.t1;
The ACCESS_HISTORY view records the get_product function in the:
DIRECT_OBJECTS_ACCESSED column because the function is explicitly named in the SQL statement, and
OBJECTS_MODIFIED column in the
directSourcesarray because the function the source for the values that are inserted into the columns.
Similarly, the table t1 is recorded in these same columns:
DIRECT_OBJECTS_ACCESSED
OBJECTS_MODIFIED
Example: Call a Stored Procedure¶
Consider the following stored procedure and assume it is stored in the schema named mydb.procedures:
create or replace procedure get_id_value(name string) returns string not null language javascript as $$ var my_sql_command = "select id from A where name = '" + NAME + "'"; var statement = snowflake.createStatement( {sqlText: my_sql_command} ); var result = statement.execute(); result.next(); return result.getColumnValue(1); $$ ;
Calling my_procedure directly results in recording the procedure details in both the
DIRECT_OBJECTS_ACCESSED and BASE_OBJECTS_ACCESSED columns as follows:
[ { "objectDomain": "PROCEDURE", "objectName": "MYDB.PROCEDURES.GET_ID_VALUE", "argumentSignature": "(NAME STRING)", "dataType": "STRING" } ]
This example is analogous to calling a UDF (in this topic).