Details
Description
We handle initializing a new instance by calling a utility method on each of the metadata and root tables. That utility method ignores the passed table id and instead sets the initial table configs for each of the metadata and root tables.
That means we end up setting them twice. I believe the action is idempotent (since we're using the same values each time), so no failure should result.
public static void initMetadataConfig(String tableId) throws IOException { try { Configuration conf = CachedConfiguration.getInstance(); int max = conf.getInt("dfs.replication.max", 512); // Hadoop 0.23 switched the min value configuration name int min = Math.max(conf.getInt("dfs.replication.min", 1), conf.getInt("dfs.namenode.replication.min", 1)); if (max < 5) setMetadataReplication(max, "max"); if (min > 5) setMetadataReplication(min, "min"); for (Entry<String,String> entry : initialMetadataConf.entrySet()) { /* XXX There should only be one of these, and the table id should be the one passed in. */ if (!TablePropUtil.setTableProperty(RootTable.ID, entry.getKey(), entry.getValue())) throw new IOException("Cannot create per-table property " + entry.getKey()); if (!TablePropUtil.setTableProperty(MetadataTable.ID, entry.getKey(), entry.getValue())) throw new IOException("Cannot create per-table property " + entry.getKey()); } } catch (Exception e) { log.fatal("error talking to zookeeper", e); throw new IOException(e); } }