|Revision as of 22:14, 23 February 2006
Wayne (Talk | contribs)
← Previous diff
|Revision as of 21:22, 22 March 2006
Wayne (Talk | contribs)
change zonesync.pl to newer Szonesync.pl and fix some links
Next diff →
|Line 407:||Line 407:|
|-||Run zonesync.pl as described below.||+||Run [[Szonesync.pl]] as described below.|
|A check is to run utilities/bin/Sls /zoneName/home/srbAdmin.zoneName||A check is to run utilities/bin/Sls /zoneName/home/srbAdmin.zoneName|
|Line 444:||Line 444:|
|What other administrative tasks are required?||What other administrative tasks are required?|
|Once you have multiple zones federated together, you need to||Once you have multiple zones federated together, you need to|
|-||periodically run a script, zonesync.pl, to ingest information from||+||periodically run a script, [[Szonesync.pl]], to ingest information from|
|the remote zones. This could be done as a cron job. See the||the remote zones. This could be done as a cron job. See the|
|script for more information. This is highly configurable since||script for more information. This is highly configurable since|
|-||zones can be run in various modes (see FedMcat.html).||+||zones can be run in various modes (see [[Fed MCAT]]).|
|Also see [[Fed MCAT]].||Also see [[Fed MCAT]].|
Revision as of 21:22, 22 March 2006
(This use to be README.zones.)
This README file contains three sections; section 1.0 gives a quick summary of using the Scommands with the SRB zone implementation so that users can start using the Scommands right away. Section 2.0 gives a primer of the SRB zone implementation which gives a more detailed explanation of the design behind the Federated MCAT implementation. And section 3.0 answers some SRB Zone Administration questions.
1.0 SRB Zone implementation in Scommands
The Federated MCAT implementation allows users to access resources and data across zones. In version 3.0, most existing operations including Sput, Sget, Scp, Sbload, Sbunload, Sregister, container operations, etc are supported across zones. The only cross zones command not supported is Sreplicate. For now, users should use Scp to copy data across zones.
Most of the modifications in the Scommands for the Federated MCAT implementation involve providing options or defaults for determining a zone for the command and the user's home zone.
The user's home zone can be specified in the .MdasEnv file using the keyword 'mcatZone' or using the env variable mcatZone. If these methods were not used to specify the user's home zone, the srbServer will query the MCAT for it which causes the srbConnect() call to be a bit slower.
In addition to the user's home zone, most operations can involve certain interzone operations. The zoneName of an operation is determined in three ways:
1) Implicitly specified by the SRB paths involved in the commands - As described in more detail below, the first entry in a SRB path specifies the zoneName where the metadata is stored. e.g., for a SRB file with path /z1/x/y/z, z1 is the zoneName of the file. Therefore, for operations involving files and collections, the zoneName of the operation is implicitly specified by the SRB paths.
"Sput -S s1 foo /z1/x/y/z" - upload the local file foo to SRB and create foo in mcat Zone 'z1', in resource 's1' of 'z1'.
"Sget /z1/x/y/z/foo" - download the SRB file foo stored in zone 'z1' to the local file system.
"Scp -S s2 /z1/x/y/z/foo /z2/a/b/c" - copy the SRB file foo in zone 'z1' to the resource 's2' of zone 'z2' and register the new file /z2/a/b/c with mcat 'z2'.
2) Explicitly specified with the -z option - Some Scommands do not involve any SRB path, so one needs to use a new -z option. The -z zoneName option explicitly specifies a zoneName.
"SgetR -z z1" - list the resources in zone 'z1'
"Smkcont -z z1 -S s1 cont1" - make a container named 'cont1' in zone 'z1'
3) Use the cwd (current working directory) to implicitly specify the zone - if the -z option is not used, the Scommands will try to get the zone through the cwd of the Scommand session.
For example, if the current working directory is /z2/x/y/z, then Slscont will list all containers belonging to the user in zone 'z2'.
NOTE: If the SRB path begins with /home or /container, the file or collection is assumed to be stored in the local zone and all commands will still work when the user is connected to the local zone. But these file/collecions will no longer be visible from a foreign zone.
2.0 SRB Zone - A Primer
What is an SRB Zone?
An SRB Zone (or zone for short) is a set of SRB servers 'brokered' or administered through a single MCAT. Hence a zone consists of one or more SRB servers along with one MCAT-enabled server. Any existing SRB system (version 2.x.x and below) can be viewed as an SRB zone.
Why are there SRB Zones?
The idea is to allow both local control and at the same time broad-area collaboration. With previous SRB versions, collaboration is done via one MCAT for the virtual organization. But even if a site has local data (a non-MCAT SRB server) and local users, their SRB MCAT interactions would still go to a central MCAT, which could be very remote. This slows the interactions and makes the local site dependent on a remote site (e.g. they can't run SRB operations during preventive maintenance at the MCAT site). While this is acceptable for many, it can be a problem for some SRB projects. The solution is the federated SRB, where multiple MCAT system can collaborate, and also function independently.
Is MCAT Zone a purely administrative concept?
A MCAT zone is both administrative and physical - physical in the sense that each MCAT zone has exactly one MCAT. It is also administrative as it gives a autonomous administrative features to each zone. One zone can join a federation of other zones only if they agree to do so.
What is the idea in defining a zone?
With a zone-based implementation, one can now view individual SRB system as a zone and SRB Version 3.0 (and above) will allow for federating across zones.
What are the main emphasis on for designing this zoned-SRB system? Is it Improving reliability, search effectiveness and/or storage capacity of MCAT metadata?
In the order of importance 1) Performance - large latency for a worldwide grid with a single MCAT 2) Local control - organizations want to have more control on the data and resources that they are share worldwide. 3) Scalability of the MCAT.
In the zone design, is it only for replicating MCAT data or can it also be used for fragmenting MCAT metadata and linking them between federated MCATs?
The first version of the federated MCAT design is going to be a very basic design. Each MCAT zone can operate entirely independently from other zone. Most MCAT metadata is not replicated except for the zone, user and user group info. The basic functionality provided by this design is to allow a user in one zone to access data and use resources in other zones in the federation. In the first version, we are providing basic zone-to-zone access mechanism. In later development, we plan to make it more transparent to use resources, build collection, manage data and apply methods seamlessly across zones.
Can you briefly give an insight to the federation algorithm that you have devised for zone federation?
It is difficult to describe the federation algorithm. Maybe an example is the best way to do it. Suppose there are 2 zones, A and B in the federation. A user in zone A wants to copy a file /A/home/john.sdsc/foo in zone A to a collection /B/home/john.sdsc/foo1 in zone B in the resource RB in zone B. She will issue the following Scommand to a server in zone A: Scd /A/home/john.sdsc Scp -S RB foo /B/home/john.sdsc/foo1 Then, the SRB servers will do the following steps: 1) The server in zone A which we will call server A, upon receiving the user request, parses the source and destination file names. 2) With the /A root path, it knows the source path is a zone A file so it queries the zone A MCAT for the needed metadata for this file (file location, file type, access control, etc). 3) With the /B root path, it knows the destination path is a zone B path which implies resource RB is a zone B resource. so it queries the zone B MCAT for the needed metadata for source RB (the host address of the resource, type of resource, etc). 4) Once these info are known, server A will open the source file in its own zone and remotely create the destination path. Then it can start moving data. 5) When the transfer is done, server A will do a remote registration of the file just created with MCAT B.
Can users connect to remote zones directly?
They can but shouldn't. If they do, they can operate OK within that zone but will not be able to do cross-zone data movement. If they first connect to their own local zone server, then all zone operations will work fine.
How does one identify a zone?
There is a zone-name given to each SRB zone. This is a char-string which identifies uniquely a zone.
How do I name my zone? How can I be sure that no other zone exists with the same name?
There is a Zone Authority which can be queried to see if a name is already in use. We also ask every SRB administrator to register their SRB zone-name with the Zone Authority. The Zone Authority is available at http://www.npaci.edu/dice/srb/zoneAuthority.html
Are there user name restrictions?
In the current version, it is assumed that there are no user name (name@domain) collisions between zones. For example, there will never be a jane@sdsc from zone A and also from zone B. So, you need to keep the domain names unique too.
What do you mean by federating across zones?
In a SRB federation there will be multiple zones (each with their own MCAT administration). In such a set up, one can register users across zones (i.e, in multiple MCATs) and these users can access data across zones (i.e.e, access data from two or more zones). Hence a federation provides a means for SRB systems (zones) to allow interactions between the zones without foregoing their autonomous nature.
How does a SRB system as it is currently run (Versions 2.x.x and below) different from the zone-system advocated by the SRB Versions 3 and above.
The old SRB systems (of one MCAT and multiple resource servers) is the definition of a zone. The main difference between the old system and the new one is that the new one allows for interactions between two such systems whereas in the old system no interaction is allowed and also if one wants to copy files between them, one needs to do that outside the SRB (eg. do an Sget from one SRB system and do an Sput into the other system).
Are all new SRB MCAT installations also Zones?
Yes. Starting with SRB 3.0, each SRB-MCAT installation will also be a Zone. You can start with it as a non-federated SRB system and then later, if you wish, federate with any of the other MCATs. For existing SRB systems, they can run as is but will need to have a zone-name when they upgrade to 3.0.
Does one zone know about users in another zone?
Yes. If two SRB zones federate, then they will exchange their user identifications. Hence one zone knows about users in the other zone.
Does a user have a home zone?
Yes. Every SRB user has one and only one home zone. They are aliens in other zones. (no dual citizenship in SRB zones)
Does it mean that my certificate or passwords get copied in multiple zones?
No. All sensitive information about a user is kept only in the home zone.
How do zones exchange user information?
There will be zone-user information exchange routines which can be launched periodically to do this.
Does one zone know about a srbObject in another zone?
Yes, provided a user uses /zone-name/home/..... to access the file.
What is the authentication mechanism used between zones?
For starters, we need to give a bit of background on how authentication and authorization work in the current single MCAT system. Currently, there can be any number of SRB data servers run by SRB users with "admin user" privileges. A normal SRB user can log onto any SRB server using his/her own credential and will be able to access data residing on any servers in the system. To be able to do this, we have a fairly robust set of server-server operations including all the basic posix I/O operations, third party transfer, server-directed parallel I/O, etc. For server-server operations, the initiating server authenticates itself using its "admin user" credential and issue requests on behalf of the normal user. In a nutshell, the only difference in privilege between a normal user and an admin user is admin users can perform operation on behalf of other users but normal users cannot. The implication though is admin users can do everything, including reading and modifying other user's data.
Hence, in a zone-based system, we need to be very careful about extending the "admin user" privilege across zones. A zone probably does not want any foreign admin user to have full admin privilege in its zone , i.e., don't want foreign admin users to be able to perform task on behalf of all users. So, we limit the privilege of a foreign admin user to be able to perform task on behalf of only users from its own zone. But the implication is if a normal user logon to a foreign zone, he/she can access only data and resources in this zone. To access data and resources in all zones, the normal user has to logon through their own zone. Briefly, a user logged into his/her home zone
What operations can a user perform, when connected to one zone, on an srbObject in another zone?
The user can perform copy, get, put, posix type functions such as open, close, read, write, and metadata access. These translate to Sput, Sget, SgetD, Smeta, Sufmeta.
Can a user put a file into another zone?
Can one zone talk to another zone which is using a different port number?
In the current implementation we expect all federating zones to use the same port numbers. We will relax this restriction in a future release.
How can a user know about resources, data, collections or containers in another zone?
Many of the Scommands (eg. SgetR, SgetD, SgetColl, etc) have been augmented with a -z option which can be used to query across zones.
A zone contains information about Domains, Resources and Metadata of srbObjects, list of administrators within the Zone etc. Does it contain other information too. Could you also briefly explain how is this information structured within a zone?
A MCAT zone has a fully functional MCAT which contains all metadata of everything that it manages locally including Domains, Resources and Metadata of srbObjects, list of administrators within the Zone. In addition, in a federated environment, it also contains information on other zones (a zone table which contains info such as zoneName, the address of the MCAT enabled server in that zone, etc), external user and user group info from other zones , etc.
Also where is the information of administrators account is stored within a zone? Is there a special kind of administrator user-type?
Administrators info is stored in the user table. They are just a special kind of user. Sensitive info such as passwords are only stored locally for local administrators.
When you federate metadata stored on MCAT server, do all information such as Domains, Resources, Administrators and File description get equally federated or can they be selectively federated? E.g.. replication/federation of only File Description metadata portion of MCAT metadata.
They are selectively federated, not by functionality as you suggested, but by data and resource controlled by the zone.
They are selectively federated, not by functionality as you suggested, but by data and resource controlled by the zone.
Can a user register a srbObject in one zone in another zone?
Can a user register a Collection in one zone in another zone?
Can a user see user-defined metadata for a srbObject across zones?
Can I register a resource in more than one zone?
Is a zone-name attached to each srbObject?
Not in the initial phases of the implementation. We plan to do that at a later stage as we add more features to the zone system.
Can I use the zone-SRB to provide scalability for MCATs?
This is one of the design goals for the zone-SRB. One should be able use multiple MCATs and load-balance them by having them run at geographically different locations and distributing users across them in different home-zones.
I have a MCAT which is getting awfully big, how can the zone-SRB help me?
This is a a good application for the zone-SRB. What you can do in this case is "semi-retire" the old MCAT (i.e., no more additions are allowed except querying) and start using a new MCAT for ingestion/registration of new srbObjects and collections. With this setup, one can access the old collection and srbObjects through the new MCAT (with some delay) but get access to new srbObjects and collections faster than would have been possible with a single large MCAT. Hence, a large system, might "retire" an MCAT every year and start using a new MCAT with good level of performance.
How can I get some information about zones?
Two of the Scommands have been extended to provide Zone information. Stoken Zone [value] will provide metadata about Zones. SgetU -Z <user> will show the Zone Name for a user.
The Java admin tool (under MCAT/java) can also be used to display Zone information.
3.0 SRB Zone - Administration
(Administrators will also need to be familiar with the other sections of this document.)
Who can ingest a new Zone name or modify Zone information? Only a SRB administrator can do that.
How can a SRB administrator perform these tasks? If the SRB Administrator is using the Scommands interface, then she can do this using the Szone command. Using this command, an SRB administrator can ingest a new zone (both local and remote), modify the metadata about existing zones, change or insert a zone for a user, modify all the metadata about a zone.
The Java admin tool (under MCAT/java) can also be used to more easily perform these tasks via its GUI. It can be used to create zones, modify zones, display zone information (local and remote, overview and in detail), and delete zones. There is also an option to modify modify a user's zone (under modify user). New users are ingested as local-zone users, but the remote admin account will need to be modified to be of the remote zone.
How does one set up a federation of zones? You use either the admin command-line utilities (Szone for most functions, and 'Stoken Zone' to list Zones) or the Java admin tool to do this (see the new set of Zone operations and new Zone option under modify user).
As of SRB 3.3, there is a new script, zoneingest.pl, (under the MCAT directory in the release) that will do steps 2 through 6 for you using as input the 'Stoken Zone' output from the remote zone. See the script (and the paragraph below) for more information.
If two SRB zones are to be fully federated you will need to do these steps once on your zone and the remote administrator will need to do them once on his/her zone (see 'Repeat 1-6 on the remote zone.' paragraph below). If you are using zoneingest.pl this 'utilities/bin/Stoken Zone > zoneFilename' will be run on each zone and the resulting zoneFilename files will need to be exchanged (physically copied) between the two SRB zone hosts.
Or, if you are using the install.pl script for testing, it can perform these steps (1-6) for you.
In the example below, the remote zone is 'B'.
- Create a local zone with a name of your choosing. If you've installed a new MCAT or if you've upgraded an existing MCAT to 3.0, a local zone called 'demozone' will exist. You can rename it to your zone name. You may also wish to update some of the other attributes of your zone.
- Add a Domain matching the one used by the admin at B.
- Add the Zone B admin user in the B domain (e.g. srbAdmin@B), as a sysadmin user type; use an arbitrary (and long) password, as their real password is in the MCAT at B.
- Add a Location for the remote zone host B
- Add a Zone (of type remote) for B
- Change the zone for the B admin user to zone B. This will give the admin user limited access to the local MCAT; they will be able to read it to synchronize.
Repeat 1-6 on the remote zone. The zone B administrator will need to perform these steps for your zone on the B MCAT. That is, you do steps 1 thru 6 to include B's zone in your zone, and the B administrator does steps 1 thru 6 to include your zone in his/her B zone.
If your Zone is to be real (i.e. not just for testing), you need to register it with the Zone Authority (as described in an above section).
Run Szonesync.pl as described below.
A check is to run utilities/bin/Sls /zoneName/home/srbAdmin.zoneName which will list the contents of the remote zone.
Must all zones in the federation use the same port number ?
No. All servers within a zone must use the same port number, but each zone in the federation can have their own port number. For each server, the port number of its own zone is determined by the "srbPort" parameter defined in the .MdasEnv file or in the "runsrb" script. The port number of foreign (external) zone is determined by the port number metadata associated with the zone in MCAT. These zones metadata is created by sys admin using the "create zone" tool.
How does one upgrade an existing MCAT to version 3.0?
Like other upgrades, use one of the scripts provided in the MCAT/data directory (such as 212to300patch.psg), using it as the standard-input for the command-line sql utility, for example 'psql < 212to300patch.psg'.
Once that completes, your local zone will be called "demozone".
You can operate a 3.0 system like this if you wish. That is, you can run the current SRB software but without doing inter-zone operations. You'll have a zone called "demozone" which will be largely unnoticed.
If you then want to make use of the zone capabilities (form a federation with other zones), you need to perform the steps outlined above under "How does one set up a federation of zones?".
What other administrative tasks are required? Once you have multiple zones federated together, you need to periodically run a script, Szonesync.pl, to ingest information from the remote zones. This could be done as a cron job. See the script for more information. This is highly configurable since zones can be run in various modes (see Fed MCAT).
Also see Fed MCAT.
See the Zones Testing writeup for a description of how one might test SRB zones and more information on how zones should operate.