[
https://issues.apache.org/jira/browse/HADOOP-15663?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16592294#comment-16592294
]
Da Zhou edited comment on HADOOP-15663 at 8/24/18 11:19 PM:
------------------------------------------------------------
Attaching patch: HADOOP-15663-HADOOP-15407-001.patch :
These config properties changes are for *TEST* purpose only, it won't affect
the production code.
- Merged ABFS and WASB configuration files into a single file.
- Removed *"fs.azure.test.account.name" *as it is a duplicate entry of
*"fs.azure.test.account.name"*
- Added new test config property *"fs.azure.wasb.account.name" *and
*"fs.azure.abfs.account.name"* to solve the properties overlap issues.
Meanwhile, the original property "fs.azure.account.name" is still supported
when running only ABFS tests or WASB tests.
- Added accountName/key verification to catch misconfigurations early.
- Added new test enable/disable control property
*"fs.azure.wasb.tests.enabled" *and *"fs.azure.abfs.tests.enabled"*. I was
trying to enable/disable tests by checking if wasb/abfs test accoutName is
missing, however there are many tests that are not depends on account name, in
case that people are trying to run those tests, I gave up that approach,
instead I added these two property which make it more clear to developers.
- Removed the ABFS emulator config, added* "fs.azure.abfs.endpoint"* to
support endpoint in the format of *IP: PORT*
- Some tests were set to ran in sequential, which is not necessary, Updated
the pom to run them in parallel.
PS: the account name was used as suffix for credentials setting properties, I
preserved this because it is helpful to avoid mis-configurations when:
- there are multiple accounts in config files
- and dev is switching accounts by comment/uncomment other accounts.
Also in case we need to support multiple accounts in future, I preserved this
suffix.
was (Author: danielzhou):
Attaching patch: HADOOP-15663-HADOOP-15407-001.patch :
These config properties changes are for *TEST* purpose only, it won't affect
the production code.
- Merged ABFS and WASB configuration files into a single file.
- Removed *"fs.azure.test.account.name" *as it is a duplicate entry of
*"fs.azure.test.account.name"*
- Added new test config property *"fs.azure.wasb.account.name" *and
*"fs.azure.abfs.account.name"* to solve the properties overlap issues.
Meanwhile, the original property "fs.azure.account.name" is still supported
when running only ABFS tests or WASB tests.
- Added accountName/key verification to catch misconfigurations early.
- Added new test enable/disable control property
*"fs.azure.wasb.tests.enabled" *and *"fs.azure.abfs.tests.enabled"*. I was
trying to enable/disable tests by checking if wasb/abfs test accoutName is
missing, however there are many tests that are not depends on account name, in
case that people are trying to run those tests, I gave up that approach,
instead I added these two property which make it more clear to developers.
- Removed the ABFS emulator config, added* "fs.azure.abfs.endpoint"* to
support endpoint in the format of *IP: PORT*
- Some tests were set to ran in sequential, which is not necessary, Updated
the pom to run them in parallel.
> ABFS: Simplify configuration
> ----------------------------
>
> Key: HADOOP-15663
> URL: https://issues.apache.org/jira/browse/HADOOP-15663
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: fs/azure
> Reporter: Thomas Marquardt
> Assignee: Da Zhou
> Priority: Major
> Attachments: HADOOP-15663-HADOOP-15407-001.patch
>
>
> Configuration for WASB and ABFS is too complex. The current approach is to
> use four files for test configuration.
> Both WASB and ABFS have basic test configuration which is committed to the
> repo (azure-test.xml and azure-bfs-test.xml). Currently these contain the
> fs.AbstractFileSystem.[scheme].impl configuration, but otherwise are empty
> except for an include reference to a file containing the endpoint
> credentials.
> Both WASB and ABFS have endpoint credential configuration files
> (azure-auth-keys.xml and azure-bfs-auth-keys.xml). These have been added to
> .gitignore to prevent them from accidentally being submitted in a patch,
> which would leak the developers storage account credentials. These files
> contain account names, storage account keys, and service endpoints.
> There is some overlap of the configuration for WASB and ABFS, where they use
> the same property name but use different values.
> 1) Let's reduce the number of test configuration files to one, if possible.
> 2) Let's simplify the account name, key, and endpoint configuration for WASB
> and ABFS if possible, but still support the legacy way of doing it, which is
> very error prone.
> 3) Let's improve error handling, so that typos or misconfiguration are not so
> difficult to troubleshoot.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
