Page tree
Skip to end of metadata
Go to start of metadata

 

Introduction

Multi data set archiver is a tool to archive several datasets together in chunks of relatively large size. When a group of datasets is selected for archive it is verified if they are all together of proper size and then they are being stored as one big container file (tar) on the destination storage.

When unarchiving data sets from a multi data set archive the following rules are obeyed:

  • Unarchiving of data sets from different containers is possible as long as the maximum unarchiving cap specified in the plugin.properties file is not exceeded.
  • All data sets from a container are unarchived even though unarchiving has been requested only for a sub set.
  • The data sets are unarchived into a share which is marked as an unarchiving scratch share.
  • In case of not enough free space in the scratch share the oldest (defined by modification time stamp) data sets are removed from the scratch share to free space. For those data sets the archiving status is set back to ARCHIVED.

At the moment deletion from the multi data set archive is not supported. That is, DeleteFromArchiveMaintenanceTask (see Maintenance Tasks) will throw a NotImplementedException.

To test the archiver find the datasets you want to archive in openbis GUI and "add to archive".

Important technical details

The archiver requires configuration of three important entities.

  • An archive destination (e.g. on Strongbox).
  • A PostgreSQL database for mapping information (i.e. which data set is in which container file).
  • An unarchiving scratch share.

Multi dataset archiver is not compatible with other archivers. You should have all data available before configuring this archiver.

Workflows

The multi data set archiver can be configured for four different workflows. The workflow is selected by the presence/absence of the properties staging-destination and replicated-destination.

Simple workflow

None of the properties  staging-destination and replicated-destination are present.

  1. Wait for enough free space on the archive destination.
  2. Store the data set in a container file directly on the archive destination.
  3. Perform sanity check. That is, getting the container file to the local disk and compare the content with the content of all data sets in the store.
  4. Add mapping data to the PostgreSQL database.
  5. Remove data sets from the store if requested.
  6. Update archiving status for all data sets.

Staging workflow

Property staging-destination is specified but replicated-destination is not.

  1. Store the data sets in a container file in the staging folder.
  2. Wait for enough free space on the archive destination.
  3. Copy the container file from the staging folder to the archive destination.
  4. Perform sanity check.
  5. Remove container file from the staging folder.
  6. Add mapping data to the PostgreSQL database.
  7. Remove data sets from the store if requested.
  8. Update archiving status for all data sets.

Replication workflow

Property replicated-destination is specified but staging-destination is not.

  1. Wait for enough free space on the archive destination.
  2. Store the data set in a container file directly on the archive destination.
  3. Perform sanity check.
  4. Add mapping data to the PostgreSQL database.
  5. Wait until the container file has also been copied (by some external process) to a replication folder.
  6. Remove data sets from the store if requested.
  7. Update archiving status for all data sets.

Some remarks:

  • Steps 5 to 7 will be performed asynchronously from the first four steps because step 5 can take quite long. In the meantime the next archiving task can already be performed.
  • If the container file isn't replicated after some time archiving will be rolled back and scheduled again.

Staging and replication workflow

When both properties staging-destination and replicated-destination are present staging and replication workflow will be combined.

Clean up

In case archiving fails all half-baked container files have to be removed. By default this is done immediately.

But in context of tape archiving systems (e.g. Strongbox) immediate deletion might not always be possible all the time. In this case a deletion request is schedule. The request will be stored in file. It will be handled in a separate thread in regular time intervals (polling time). If deletion isn't possible after some timeout an e-mail will be sent. Such deletion request will still be handled but the e-mail allows manual intervention/deletion. Note, that deletion requests for non-existing files will always be handled successfully.

Configuration steps

  • Disable existing archivers
    • Find all properties of a form archiver.* in servers/datastore_server/etc/service.properties and remove them.
    • Find all DSS core plugins of type miscellaneous which define an archiver. Disable them by adding an empty marker file named disabled.
  • Enable archiver
    • Configure a new DSS core plugin of type miscellaneous:

      multi-dataset-archiver/1/dss/miscellaneous/archiver/plugin.properties
      archiver.class = ch.systemsx.cisd.openbis.dss.generic.server.plugins.standard.archiver.MultiDataSetArchiver
      
      # Temporary folder (needed for sanity check). Default: Value provided by Java system property java.io.tmpdir. Usually /tmp
      # archiver.temp-folder = <java temp folder>
      
      # Archive destination
      archiver.final-destination = path/to/strongbox/as/mounted/resource
      
      # Staging folder (needed for 'staging workflow' and 'staging and replication workflow')
      archiver.staging-destination = path/to/local/stage/area
      
      # Replication folder (needed for 'replication workflow' and 'staging and replication workflow')
      archiver.replicated-destination = path/to/mounted/replication/folder
      
      # The archiver will refuse to archive group of data sets, which together are smaller than this value
      archiver.minimum-container-size-in-bytes = 15000000
      
      # The archiver will refuse to archive group of data sets, which together are bigger than this value.
      # The archiver will ignore this value, when archiving single data set.
      archiver.maximum-container-size-in-bytes = 35000000
      
      # This variable is meant for another use case, than this archiver, but is shared among all archivers.
      # For this archiver it should be specified for something safely larger than maximum-container-size-in-bytes
      archiver.batch-size-in-bytes = 80000000
      
      # Archiving can be speed up if setting this flag to false (default value: true). But this works only if the data sets
      # to be archived do not contain hdf5 files which are handled as folders (like the thumbnail h5ar files in screening/microscopy).
      # archiver.hdf5-files-in-data-set = true
      
      # Whether all data sets should be archived in a top level directory of archive or with sharding (the way data sets are stored in openbis internal store)
      # archiver.with-sharding = false
      
      # Polling time for evaluating free space on archive destination
      # archiver.waiting-for-free-space-polling-time = 1 min
      
      # Maximum waiting time for free space on archive destination
      # archiver.waiting-for-free-space-time-out = 4 h
      
      # Minimum free space on archive destination after container file has been added.
      # archiver.minimum-free-space-in-MB = 1024
      
      # Polling time for waiting on replication. Only needed if archiver.replicated-destination is specified.
      # archiver.finalizer-polling-time = 1 min
      
      # Maximum waiting time for replication finished.  Only needed if archiver.replicated-destination is specified.
      # archiver.finalizer-max-waiting-time = 1 d
      
      # Maximum total size (in MB) of data sets that can be scheduled for unarchiving at any given time. When not specified, defaults to 1 TB.
      # Note also that the value specified must be consistent with the scratch share size. 
      # maximum-unarchiving-capacity-in-megabytes = 200000
      
      # Delay unarchiving. Needs MultiDataSetUnarchivingMaintenanceTask.
      # archiver.delay-unarchiving = false
      
      # Size of the buffer used for copying data. Default value: 1048576 (i.e. 1 MB). This value is only important in case of accurate
      # measurements of data transfer rates. In case of expected fast transfer rates a larger value (e.g. 10 MB) should be used.
      # archiver.buffer-size = 1048576
      
      # Maximum size of the writing queue for copying data. Reading from the data store and writing to the TAR file is 
      # done in parallel. The default value 5 * archiver.buffer-size. 
      # archiver.maximum-queue-size-in-bytes = 5242880
      
      # Path (absolute or relative to store root) of an empty file. If this file is present starting 
      # archiving will be paused until this file has been removed. 
      # This property is useful for archiving media/facilities with maintenance downtimes.
      # archiver.pause-file = pause-archiving
      
      # Time interval between two checks whether pause file still exists or not.
      # archiver.pause-file-polling-time = 10 min
      
      #-------------------------------------------------------------------------------------------------------
      # Clean up properties
      # 
      # A comma-separated list of path to folders which should be cleaned in a separate thread
      #archiver.cleaner.file-path-prefixes-for-async-deletion = <absolute path 1>, <absolute path 2>, ...
      
      # A folder which will contain deletion request files. This is a mandatory property if 
      # archiver.cleaner.file-path-prefixes-for-async-deletion is specified.
      #archiver.cleaner.deletion-requests-dir = <some local folder>
      
      # Polling time interval for looking and performing deletion requests. Default value is 10 minutes.
      #archiver.cleaner.deletion-polling-time = 10 min
      
      # Time out of deletion requests. Default value is one day.
      #archiver.cleaner.deletion-time-out = 24 h
      
      # Optional e-mail address. If specified every integer multiple of the timeout period an e-mail is send to 
      # this address listing all deletion requests older than specified timeout.
      #archiver.cleaner.email-address = <some valid e-mail address>
      
      # Optional e-mail address for the 'from' field.
      #archiver.cleaner.email-from-address = <some well-formed e-mail address>
      
      # Subject for the 'subject' field. Mandatory if an e-mail address is specified.
      #archiver.cleaner.email-subject = Deletion failure
      
      # Template with variable ${file-list} for the body text of the e-mail. The variable will be replaced by a list of
      # lines. Two lines for each deletion request. One for the absolute file path and one of the request time stamp.
      # Mandatory if an e-mail address is specified.
      #archiver.cleaner.email-template = The following files couldn't be deleted:\n${file-list}
      
      #-------------------------------------------------------------------------------------------------------
      # The following properties are necessary in combination with data source configuration
      multi-dataset-archive-database.kind = prod
      multi-dataset-archive-sql-root-folder = datastore_server/sql/multi-dataset-archive
      

      You should make sure that all destination directories exist and DSS has read/write privileges before attempting archiving (otherwise the operation will fail).
      Add the core plugin module name (here multi-dataset-archiver) to the property enabled-modules of core-plugin.properties.

  • Enable PostgreSQL datasource
    • Configure a new DSS core plugin of type data-sources:

      multi-dataset-archiver/1/dss/data-sources/multi-dataset-archiver-db/plugin.properties
      version-holder-class = ch.systemsx.cisd.openbis.dss.generic.server.plugins.standard.archiver.dataaccess.MultiDataSetArchiverDBVersionHolder
      databaseEngineCode = postgresql
      basicDatabaseName = multi_dataset_archive
      urlHostPart = ${multi-dataset-archive-database.url-host-part:localhost}
      databaseKind = ${multi-dataset-archive-database.kind:prod}
      scriptFolder = ${multi-dataset-archive-sql-root-folder:}
      owner = ${multi-dataset-archive-database.owner:}
      password = ${multi-dataset-archive-database.password:}
  • Create a share which will be used exclusively as a scratch share for unarchiving. To mark it for this purpose add the following share.properties file to the share (e.g. <mounted share>/store/1/share.properties):

    share.properties
     unarchiving-scratch-share = true
  • It is recommended to do archiving in a separate queue in order to avoid situation when fast processing plugin tasks are not processes because multi data set archiving tasks can take quite long. If one of the two workflows with replication is selected (i.e. archiver.replicated-destination) a second processing plugin (ID Archiving Finalizer) is used. It should run in a queue different from the queue used for archiving. The following setting in DSS service.properties covers all workflows:

    service.properties
    data-set-command-queue-mapping = archiving:Archiving|Copying data sets to archive, unarchiving:Unarchiving, archiving-finalizer:Archiving Finalizer

Recovery from corrupted archiving queues

In case the queues with the archiving commands get corrupted, they cannot be used any more, they need to be deleted before the DSS starts and a new one will be created. The typical scenario where this happens is when you get out of space on the disk where the queues are stored.

The following steps describe how to recover from such a situation.

  1. Finding out the data sets that are in 'ARCHIVE_PENDING' status.

    SELECT data_id, size, present_in_archive, share_id, location FROM external_data WHERE status = 'ARCHIVE_PENDING';
     
    openbis_prod=> SELECT data_id, size, present_in_archive, share_id, location FROM external_data WHERE status = 'ARCHIVE_PENDING'; 
     data_id |    size     | present_in_archive | share_id |                               location                                
    ---------+-------------+--------------------+----------+-----------------------------------------------------------------------
        3001 | 34712671864 | f                  | 1        | 585D8354-92A3-4C24-9621-F6B7063A94AC/17/65/a4/20170712111421297-37998
        3683 | 29574172672 | f                  | 1        | 585D8354-92A3-4C24-9621-F6B7063A94AC/39/6c/b0/20171106181516927-39987
        3688 | 53416316928 | f                  | 1        | 585D8354-92A3-4C24-9621-F6B7063A94AC/ca/3b/93/20171106183212074-39995
        3692 | 47547908096 | f                  | 1        | 585D8354-92A3-4C24-9621-F6B7063A94AC/b7/26/85/20171106185354378-40002
  2. The data sets found, can be or not in the archiving process. This is not easy to find out instantly. It's easier just to execute the above statement in subsequent days.

  3. If the data sets are still in 'ARCHIVE_PENDING' after a sensible amount of time (1 week for example) and there is no other issues, like the archiving destination is not available there is a good change, they are really stuck on the process.
  4. Reaching this point, the data sets are most likely still on the data store as indicated by the combination of share ID and location indicated. Verify this! If they are not there hope they are archived or you are on trouble.
  5. If they are on the store, you need to set the status to available again using a SQL statement.

     openbis_prod=> UPDATE external_data SET status = 'AVAILABLE', present_in_archive = 'f'  WHERE data_id IN (SELECT id FROM data where code in ('20170712111421297-37998', '20171106181516927-39987')); 

     

    If there is half copied files on the archive destination, these need to be delete too, to find them run the next queries.


    # To find out the containers:
     
    SELECT * FROM data_sets WHERE CODE IN('20170712111421297-37998', '20171106181516927-39987', '20171106183212074-39995', '20171106185354378-40002');
    
    multi_dataset_archive_prod=> SELECT * FROM data_sets WHERE CODE IN('20170712111421297-37998', '20171106181516927-39987', '20171106183212074-39995', '20171106185354378-40002');
     id  |          code           | ctnr_id | size_in_bytes 
    -----+-------------------------+---------+---------------
     294 | 20170712111421297-37998 |      60 |   34712671864
     295 | 20171106185354378-40002 |      61 |   47547908096
     296 | 20171106183212074-39995 |      61 |   53416316928
     297 | 20171106181516927-39987 |      61 |   29574172672
    (4 rows)
    
    multi_dataset_archive_prod=> SELECT * FROM containers WHERE id IN(60, 61);
     id |                    path                     | unarchiving_requested 
    ----+---------------------------------------------+-----------------------
     60 | 20170712111421297-37998-20171108-105339.tar | f
     61 | 20171106185354378-40002-20171108-130342.tar | f
     
    

    NOTE: We have never seen it but if there is a container with data sets in different archiving status then, you need to recover the ARCHIVED data sets from the container and copy them manually to the data store before being able to continue.

    multi_dataset_archive_prod=> SELECT * FROM data_sets WHERE ctnr_id IN(SELECT ctnr_id FROM data_sets WHERE CODE IN('20170712111421297-37998', '20171106181516927-39987', '20171106183212074-39995', '20171106185354378-40002'));
  6. After deleting the files clean up the multi dataset archiver database.

    openbis_prod=> DELETE FROM containers WHERE id IN (SELECT ctnr_id FROM data_sets WHERE CODE IN('20170712111421297-37998', '20171106181516927-39987', '20171106183212074-39995', '20171106185354378-40002'));
    
  • No labels