Synapse REST API Auditing and Logging

Introduction

Auditing data for the Synapse REST API is captured by a Spring Interceptor: AccessInterceptor that is similar to web filter.  This interceptor is configured to listen to all web services calls made to the repository services.  For each call, the AccessInterceptor will gather data to fill out an AccessRecord model object.  The AccessRecord data is then written as zipped CSV files directly to the  prod.access.record.sagebase.org  S3 bucket.  These CSV files are initially too small to process efficiently so a worker process merges the files by hour.

AccessRecord S3 Files

All AccessRecord CSV data for a single hour from all EC2 instance of a stack are into a single file.  The following is an example of the resulting path:

https://s3.amazonaws.com/prod.access.record.sagebase.org/000000013/2013-09-17/14-39-03-055-75ed6416-438d-4688-8789-3df56f4e4670.csv.gz

The above path is composed of the following parts:

https://s3.amazonaws.com/prod.access.record.sagebase.org/<instance_number>/<year_month_day>/<hour_minutes_seconds_miliseconds>-<UUID>.csv.gz

Here is some sample data from one of the access record files:

Column Description

• returnedObjectId - For any method that returns an object with an ID, this column will contain the returned ID.  This is the only way to determine the ID of a newly created object from a POST.
• elaseMS - The elapse time of the call in milliseconds.
• timestamp - The exact time the call was made in epoch time (milliseconds since 1/1/1970).
• via - The value of the "via" header (see: http://en.wikipedia.org/wiki/List_of_HTTP_header_fields).
• host - The value of the "host" header (see: http://en.wikipedia.org/wiki/List_of_HTTP_header_fields).
• threadId- The ID of the thread used to process the request.
• userAgent - The value of the "User-Agent" header (see: http://en.wikipedia.org/wiki/List_of_HTTP_header_fields).
• queryString - The value of the request queryString.
• sessionId - For each call a new UUID is generated for the sessionId.  The sessionId is also bound to the logging thread context and written in all log entries.  This ties access records to log entries.
• xForwardedFor - The value of the "X-Forwarded-For" header (see: http://en.wikipedia.org/wiki/List_of_HTTP_header_fields).
• resquestURL - The URL of the request.
• userID - For calls where the users is authenticated via a sessionToken or an API key, this column will contain the numeric ID of the user.
• origin - The value of the "Origin" header (see: http://en.wikipedia.org/wiki/List_of_HTTP_header_fields).
• date - The year-month-day date string.
• method - The HTTP method: GET, POST, PUT, DELETE
• vmId - When each EC2 instances of a stack starts, a new unique identifier for the JVM is issued.  This is captured in the access log so calls form a single machine can be grouped together.
• instance - The instance number of the stack.
• stack - The stack identifier.  This will always be "prod" for production stacks.
• success - Set to true when a call complete without an exception, otherwise set to false.  The stack trace of exceptions can be found by searching the logs for the the sessionId of any failed access records.

Audit Analysis

Analysis of the audit data can be done using EMR-Hive.  The AWS extensions to Hive include support for reading zipped CSV data directly from S3.  This means we can launch an Elastic Map Reduce (EMR) Hive cluster and copy all access record data from S3 to the Hive cluster.  Once the data is loaded on the cluster add-hock and canned quires can be executed to generate reports or discover new trends.  The following outlines how to get started with the analysis of the Synapse Audit data.

Launch a EMR Hive Cluster

Launch a new EMR Hive cluster (see Launch a Hive Cluster for more information).

Log into the Production AWS account and navigate to the Elastic MapReduce Service page and select "Create New Job Flow":

Select a Hive Program:

Choose "Start an Interactive Hive Session"

Choose the number and types of the EC2 instances for the cluster. There must be at least one "Master Instance Group" instance and two "Core Instance Group" instances.

Setup the Amazon EC2 Key Pair to "prod-key-pair".  You will then need to use the prod-key-pair to SSH into the cluster.

No bootstrap is needed:

Final check for the cluster configuration.  Select "Create Job Flow" to launch the cluster:

Once the cluster is launched and active you can find the "Master Public DNS Name" needed to SSH into the cluster:

SSH into Hive Cluster

Once the cluster is active SSH into the master node using the prod-key-pair (see: AccessingInstancesLinux for more information).  Note: the user name must be "hadoop":

ssh -i prod-key-pair.pem hadoop@ec2-54-242-184-13.compute-1.amazonaws.com

Once connected to the Hive master node, start a new interactive hive session by typing "hive"

hadoop@ip-10-28-72-37:~$ hive

Creating Hive Tables

Once you have an interactive Hive session with the master node of the cluster you are ready to setup the tables that will be used for analysis.  First we must create the external table for access record CSV data in S3:

CREATE EXTERNAL TABLE access_record_s3 ( 
returnObjectId string,
elapseMS int,
timestamp int,
via string,
host string,
threadId int,
userAgent string,
queryString string,
sessionId string,
xForwardedFor string,
requestURL string,
userId int,
origin string,
date string,
method string,
vmId string,
instance string,
stack string,
success string
 )
PARTITIONED BY ( datep string )
ROW FORMAT DELIMITED FIELDS TERMINATED BY ',' LINES TERMINATED BY '\n'
LOCATION 's3://prod.access.record.sagebase.org/000000012';

In this example,  we created a table using prod-12 data as the source for the table (see: line:24 LOCATION 's3://prod.access.record.sagebase.org/000000009').  Make sure to set the location to the stack data you want to analyses.

At this point our table 'access_record_s3' exists but it does not contain any data.  This is because we must manually add a partition for each date we want to load into the table.  By default this is no data in a partitioned table.  Lets add some partitions to this external table:

ALTER TABLE access_record_s3 ADD PARTITION(datep='2013-09-15') location 's3://prod.access.record.sagebase.org/000000012/2013-09-15';
ALTER TABLE access_record_s3 ADD PARTITION(datep='2013-09-16') location 's3://prod.access.record.sagebase.org/000000012/2013-09-16';
ALTER TABLE access_record_s3 ADD PARTITION(datep='2013-09-17') location 's3://prod.access.record.sagebase.org/000000012/2013-09-17';

Note: Adding the current day's data as a partition will be problematic. The files for the current day are in flux as the worker will constantly merges the small files into larger files. This will cause any hive queries to fail.

The above command will add all access record data from 2013-09-15 through 2013-09-17 to our external table 'access_record_s3'.  We can now run a query to validate there is data in our external table:

select count(*) from access_record_s3;

The above query produced the following:

MapReduce Total cumulative CPU time: 1 minutes 18 seconds 600 msec
Ended Job = job_201309172347_0003
Counters:
MapReduce Jobs Launched:
Job 0: Map: 72  Reduce: 1   Cumulative CPU: 78.6 sec   HDFS Read: 19440 HDFS Write: 8 SUCCESS
Total MapReduce CPU Time Spent: 1 minutes 18 seconds 600 msec
OK
1833436
Time taken: 175.089 seconds, Fetched: 1 row(s)
hive>

We now have an external table configured with three days worth of access data.  However, any query we run against this external table will be slow as Hive must download all of the data from S3 each time a query runs.  To speed things up we can copy all of the data from our external table into a table located directly on the EMR Hive cluster.  First create the local table:

CREATE TABLE access_record_local ( 
returnObjectId string,
elapseMS int,
timestamp int,
via string,
host string,
threadId int,
userAgent string,
queryString string,
sessionId string,
xForwardedFor string,
requestURL string,
userId int,
origin string,
date string,
method string,
vmId string,
instance string,
stack string,
success string
 )
ROW FORMAT DELIMITED FIELDS TERMINATED BY ',' LINES TERMINATED BY '\n';

Now copy all data from the external table into the local table:

insert overwrite table access_record_local select 
returnObjectId,
elapseMS,
timestamp,
via,
host,
threadId,
userAgent,
queryString,
sessionId,
xForwardedFor,
requestURL,
userId,
origin,
date,
method,
vmId,
instance,
stack,
success from access_record_s3 where datep > '2013-09-14';