Github user ctubbsii commented on a diff in the pull request:

    https://github.com/apache/accumulo/pull/253#discussion_r129890129
  
    --- Diff: 
server/base/src/main/java/org/apache/accumulo/server/fs/PerTableVolumeChooser.java
 ---
    @@ -18,61 +18,146 @@
     
     import java.util.concurrent.ConcurrentHashMap;
     
    -import org.apache.accumulo.core.client.impl.Table;
    +import org.apache.accumulo.core.client.AccumuloException;
    +import org.apache.accumulo.core.conf.AccumuloConfiguration;
     import org.apache.accumulo.core.conf.Property;
     import org.apache.accumulo.server.client.HdfsZooInstance;
     import org.apache.accumulo.server.conf.ServerConfigurationFactory;
     import org.apache.accumulo.server.conf.TableConfiguration;
    +import org.apache.accumulo.start.classloader.vfs.AccumuloVFSClassLoader;
    +import org.slf4j.Logger;
    +import org.slf4j.LoggerFactory;
     
     /**
    - * A {@link VolumeChooser} that delegates to another volume chooser based 
on the presence of an experimental table property,
    - * {@link Property#TABLE_VOLUME_CHOOSER}. If it isn't found, defaults back 
to {@link RandomVolumeChooser}.
    + * A {@link VolumeChooser} that delegates to another volume chooser based 
on other properties: table.custom.volume.chooser for tables, and
    + * general.custom.scoped.volume.chooser for scopes. 
general.custor.{scope}.volume.chooser can override the system wide setting for
    + * general.custom.scoped.volume.chooser. At the this this was written, the 
only known scope was "logger".
      */
     public class PerTableVolumeChooser implements VolumeChooser {
    -
    -  private final VolumeChooser fallbackVolumeChooser = new 
RandomVolumeChooser();
    +  private static final Logger log = 
LoggerFactory.getLogger(PerTableVolumeChooser.class);
       // TODO Add hint of expected size to construction, see ACCUMULO-3410
       /* Track VolumeChooser instances so they can keep state. */
    -  private final ConcurrentHashMap<Table.ID,VolumeChooser> 
tableSpecificChooser = new ConcurrentHashMap<>();
    +  private final ConcurrentHashMap<String,VolumeChooser> 
tableSpecificChooser = new ConcurrentHashMap<>();
    +  private final ConcurrentHashMap<String,VolumeChooser> 
scopeSpecificChooser = new ConcurrentHashMap<>();
    +  private final RandomVolumeChooser randomChooser = new 
RandomVolumeChooser();
    +
       // TODO has to be lazily initialized currently because of the reliance 
on HdfsZooInstance. see ACCUMULO-3411
       private volatile ServerConfigurationFactory serverConfs;
     
    +  public static final String TABLE_VOLUME_CHOOSER = 
Property.TABLE_ARBITRARY_PROP_PREFIX.getKey() + "volume.chooser";
    +
    +  public static final String SCOPED_VOLUME_CHOOSER(String scope) {
    +    return Property.GENERAL_ARBITRARY_PROP_PREFIX.getKey() + scope + 
".volume.chooser";
    +  }
    +
    +  public static final String DEFAULT_SCOPED_VOLUME_CHOOSER = 
SCOPED_VOLUME_CHOOSER("scoped");
    +
       @Override
    -  public String choose(VolumeChooserEnvironment env, String[] options) {
    -    VolumeChooser chooser = null;
    -    if (env.hasTableId()) {
    -      // This local variable is an intentional component of the 
single-check idiom.
    -      ServerConfigurationFactory localConf = serverConfs;
    -      if (localConf == null) {
    -        // If we're under contention when first getting here we'll throw 
away some initializations.
    -        localConf = new 
ServerConfigurationFactory(HdfsZooInstance.getInstance());
    -        serverConfs = localConf;
    +  public String choose(VolumeChooserEnvironment env, String[] options) 
throws AccumuloException {
    +    if (log.isTraceEnabled()) {
    +      log.trace("PerTableVolumeChooser.choose");
    +    }
    +    VolumeChooser chooser;
    +    if (!env.hasTableId() && !env.hasScope()) {
    +      // Should only get here during Initialize. Configurations are not 
yet available.
    +      return randomChooser.choose(env, options);
    --- End diff --
    
    This is a weird case for the per-table chooser to be forced to handle. It 
seems "init" should be a scope instead. Random should never just be used.


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---

Reply via email to