Package edu.brown.designer.partitioners.plan

Examples of edu.brown.designer.partitioners.plan.TableEntry

63
64
65
66
67
68
69
70
71
72
73
74
75
        assertNotNull(pplan);
        assertEquals(catalogContext.getDataTables().size(), pplan.getTableEntries().size());

        for (Table catalog_tbl : pplan.getTableEntries().keySet()) {
            if (catalog_tbl.getSystable()) continue;
            TableEntry pentry = pplan.getTableEntries().get(catalog_tbl);
            assertNotNull("Null PartitionEntry for " + catalog_tbl, pentry);
            Collection<Column> pkey_columns = CatalogUtil.getPrimaryKeyColumns(catalog_tbl);
            PartitionMethodType expected = (pkey_columns.isEmpty() ? PartitionMethodType.REPLICATION : PartitionMethodType.HASH);
            assertEquals("Invalid PartitionMethodType for " + catalog_tbl, expected, pentry.getMethod());
        } // FOR
        System.err.println(pplan);
    }
View Full Code Here
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
            public int getRowCount() { return (tables.size()); }
            public Object getValueAt(int row, int col) {
                String ret = null;
               
                Table catalog_tbl = tables.get(row);
                TableEntry entry = plan.getTableEntries().get(catalog_tbl);
                switch (col) {
                    case 0:
                        ret = catalog_tbl.getName();
                        break;
                    case 1:
                        ret = entry.getMethod().toString();
                        break;
                    case 2:
                        ret = (entry.getAttribute() != null ? entry.getAttribute().getName() : "-");
                        break;
                    case 3:
                        ret = (entry.getParent() != null ? entry.getParent().getName() : "-");
                        break;
                    case 4:
                        ret = (entry.getParentAttribute() != null ? entry.getParentAttribute().getName() : "-");
                        break;
                } // SWITCH
                return (ret);
            }
            public boolean isCellEditable(int row, int col) {
View Full Code Here
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
            LOG.debug("Selecting partitioning Column for " + this.info.catalogContext.database.getTables().size() + " Tables");
        double total_memory_used = 0.0;
        boolean calculate_memory = (hints.force_replication_size_limit != null && hints.max_memory_per_partition != 0);
        for (Table catalog_tbl : CatalogUtil.getDataTables(info.catalogContext.database)) {
            String table_key = CatalogKey.createKey(catalog_tbl);
            TableEntry pentry = null;
            Column col = null;
            Collection<Column> pkey_columns = CatalogUtil.getPrimaryKeyColumns(catalog_tbl);

            TableStatistics ts = info.stats.getTableStatistics(catalog_tbl);
            assert (ts != null) : "Null TableStatistics for " + catalog_tbl;
            double size_ratio = (calculate_memory ? (ts.tuple_size_total / (double) hints.max_memory_per_partition) : 0);

            // Replication
            if (hints.force_replication.contains(table_key) || (calculate_memory && ts.readonly && size_ratio <= hints.force_replication_size_limit) || pkey_columns.isEmpty()) {
                total_memory_used += size_ratio;
                if (debug.val)
                    LOG.debug("Choosing " + catalog_tbl.getName() + " for replication");
                col = ReplicatedColumn.get(catalog_tbl);
                pentry = new TableEntry(PartitionMethodType.REPLICATION, col, null, null);

                // Hash Primary Key
            } else {
                total_memory_used += (size_ratio / (double) info.getNumPartitions());

                if (hints.enable_multi_partitioning == false || pkey_columns.size() == 1) {
                    col = CollectionUtil.first(pkey_columns);
                    pentry = new TableEntry(PartitionMethodType.HASH, col, null, null);
                } else {
                    col = MultiColumn.get(pkey_columns.toArray(new Column[0]));
                    pentry = new TableEntry(PartitionMethodType.HASH, col, null, null);
                }
                assert (pentry.attribute != null) : catalog_tbl;
            }
            if (debug.val)
                LOG.debug(String.format("[%02d] %s", pplan.getTableCount(), col.fullName()));
View Full Code Here
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
                                // System.exit(1);
                                // }
                            }
                        }

                        TableEntry entry = new TableEntry(method, attribute, parent_table, parent_attribute);
                        partition_plan.getTableEntries().put((Table) element.getCatalogItem(), entry);
                        return;
                    }
                }.traverse(root);
            } // FOR roots
            LOG.info(partition_plan);
            partition_plans.put(ptree, partition_plan);
        } // FOR trees

        //
        // Now for each relation, make a tally for the different ways that it
        // could be partitioned
        // This will then be used to generate the final PartitionMapping
        //
        PartitionPlan pplan = new PartitionPlan();
        for (Table catalog_tbl : info.catalogContext.database.getTables()) {
            //
            // For each table, look at the PartitionPlan entries that we created
            // above and see
            // whether it references our table. If it does, then we need to
            // increase our count
            // by one.
            //
            Map<TableEntry, Double> counts = new HashMap<TableEntry, Double>();
            LOG.debug("Counting PartitionPlan entries for " + catalog_tbl);
            for (PartitionTree ptree : partition_plans.keySet()) {
                PartitionPlan partition_plan = partition_plans.get(ptree);
                // System.out.println("Mapping Tables: " + mapping.keySet());
                //
                // We found a partition plan that references our table, so then
                // we need to grab
                // the entry and include it in our count list. Note that the
                // PartitionPlan.Entry
                // object knows how to properly tell whether it has the same
                // attributes as
                // other entry objects, so the count should be properly updated.
                //
                if (partition_plan.getTableEntries().containsKey(catalog_tbl)) {
                    //
                    // Exclude HASH entries without attributes...
                    //
                    TableEntry entry = partition_plan.getTableEntries().get(catalog_tbl);
                    if (entry.getMethod() == PartitionMethodType.HASH && entry.getAttribute() == null) {
                        LOG.warn("Skipping entry for " + catalog_tbl + " because it does not have any partitioning attributes");
                    } else {
                        LOG.debug("Match: " + partition_plan);
                        //
                        // We need to weight this entry by the weight of the
                        // PartitionTree
                        // that it was derived from
                        //
                        double count = ptree.getWeight();
                        if (counts.containsKey(entry))
                            count += counts.get(entry);
                        counts.put(entry, count);
                    }
                } // FOR
            } // FOR

            //
            // If a table was either hashed or mapped on the same attributes,
            // then
            // always go for map
            //
            Iterator<TableEntry> it = counts.keySet().iterator();
            while (it.hasNext()) {
                TableEntry entry0 = it.next();
                if (entry0.getMethod() == PartitionMethodType.MAP)
                    continue;

                boolean remove = false;
                for (TableEntry entry1 : counts.keySet()) {
                    if (entry0 == entry1)
                        continue;
                    if (entry0.getMethod() == PartitionMethodType.HASH && entry1.getMethod() == PartitionMethodType.MAP && entry0.getAttribute().equals(entry1.getAttribute())) {
                        remove = true;
                    }
                } // FOR
                if (remove) {
                    LOG.info("Removing " + entry0 + " because a duplicate entry for a MAP already exists");
                    it.remove();
                    counts.remove(entry0);
                }
            } // WHILE

            //
            // If our counts for the current table is not empty, then we need to
            // need to pick
            // the one with the greatest count
            //
            if (!counts.isEmpty()) {
                for (TableEntry entry : counts.keySet()) {
                    LOG.debug("[" + counts.get(entry) + "]: " + entry);
                } // FOR
                  //
                  // Loop through and pick the entries with the greatest weight
                  // We use a set to warn about multiple entries that could be
                  // picked (which is another decision problem)
                  //
                Set<TableEntry> picked_entries = new HashSet<TableEntry>();
                double max_cnt = 0;
                for (TableEntry entry : counts.keySet()) {
                    double entry_cnt = counts.get(entry);
                    //
                    // If the entry's weight is the same or equal to the current
                    // max weight, then
                    // add it to our list of possible selections. Note that if
                    // it's greater than the
                    // current max weight, then we need to clear our previous
                    // entries
                    //
                    if (entry_cnt >= max_cnt) {
                        if (entry_cnt > max_cnt)
                            picked_entries.clear();
                        picked_entries.add(entry);
                        max_cnt = entry_cnt;
                    }
                } // FOR
                assert (picked_entries.isEmpty() == false);
                if (picked_entries.size() > 1) {
                    LOG.warn("Multiple entries found with the same count for " + catalog_tbl + ". Picking the first one that has a parent");
                    pplan.getTableEntries().put(catalog_tbl, CollectionUtil.first(picked_entries));
                } else {
                    // Just grab the only one and stick it in the PartitionPlan
                    pplan.getTableEntries().put(catalog_tbl, CollectionUtil.first(picked_entries));
                }
                // System.out.println(catalog_tbl + " => " +
                // final_mapping.get(catalog_tbl).toString());
                //
                // This is bad news all around...
                //
            } else {
                LOG.warn("Failed to find any PartitionPlan entries that reference " + catalog_tbl);
            }
        } // FOR tables

        //
        // HACK: Add in any tables we missed as replicated
        //
        for (Table catalog_tbl : info.catalogContext.database.getTables()) {
            if (pplan.getTableEntries().get(catalog_tbl) == null) {
                pplan.getTableEntries().put(catalog_tbl, new TableEntry(PartitionMethodType.REPLICATION, null, null, null));
            }
        } // FOR

        pplan.initializeDependencies();
        return (pplan);
View Full Code Here
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
                    return;
                }

                @Override
                protected void callback(DesignerVertex element) {
                    TableEntry entry = PartitionPlanTreeGenerator.this.pplan.getTableEntries().get((Table) element.getCatalogItem());
                    // Bad Mojo!
                    if (entry == null) {
                        LOG.warn("ERROR: No PartitionPlan entry for '" + element + "'");
                        // Non-Root
                    } else if (entry.getParent() != null) {
                        LOG.debug("Trying to create: " + entry.getParent() + "->" + element);
                        DesignerVertex parent = info.dgraph.getVertex(entry.getParent());

                        element.setAttribute(ptree, PartitionTree.VertexAttributes.ATTRIBUTE.name(), entry.getAttribute());
                        element.setAttribute(ptree, PartitionTree.VertexAttributes.METHOD.name(), entry.getMethod());

                        if (parent != null && !ptree.containsVertex(element)) {
                            if (!ptree.containsVertex(parent))
                                ptree.addVertex(parent);
                            // System.out.println("FINAL GRAPH: " + parent +
                            // "->" + element);
                            DesignerEdge edge = new DesignerEdge(ptree);
                            ptree.addEdge(edge, parent, element, EdgeType.DIRECTED);
                            element.setAttribute(ptree, PartitionTree.VertexAttributes.PARENT_ATTRIBUTE.name(), entry.getParentAttribute());
                        }
                        // Root
                    } else if (!ptree.containsVertex(element)) {
                        ptree.addVertex(element);
                        element.setAttribute(ptree, PartitionTree.VertexAttributes.ATTRIBUTE.name(), entry.getAttribute());
                        element.setAttribute(ptree, PartitionTree.VertexAttributes.METHOD.name(), entry.getMethod());
                    }
                }
            }.traverse(root);
        } // FOR
        return;
View Full Code Here
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
            Collection<Column> forced_columns = hints.getForcedTablePartitionCandidates(catalog_tbl);
            TableStatistics ts = info.stats.getTableStatistics(catalog_tbl);
            assert (ts != null) : "Null TableStatistics for " + catalog_tbl;
            double partition_size = (calculate_memory ? (ts.tuple_size_total / (double) info.getNumPartitions()) : 0);
            double partition_ratio = (calculate_memory ? (ts.tuple_size_total / (double) hints.max_memory_per_partition) : 0);
            TableEntry pentry = null;

            if (debug.val) {
                Map<String, Object> m = new ListOrderedMap<String, Object>();
                m.put("Read Only", ts.readonly);
                m.put("Table Size", StringUtil.formatSize(ts.tuple_size_total));
                m.put("Table Partition Size", StringUtil.formatSize((long)partition_size));
                m.put("Table Partition Ratio", String.format("%.02f", partition_ratio));
                m.put("Total Partition Size", String.format("%s / %s", StringUtil.formatSize(total_partitionSize), StringUtil.formatSize(hints.max_memory_per_partition)));
                m.put("Total Partition Ratio", String.format("%.02f", total_partitionRatio));
                LOG.debug(String.format("%s\n%s", catalog_tbl.getName(), StringUtil.formatMaps(m)));
            }

            // -------------------------------
            // Replication
            // -------------------------------
            if (hints.force_replication.contains(table_key) || (calculate_memory && ts.readonly && hints.enable_replication_readonly && partition_ratio <= hints.force_replication_size_limit)) {
                total_partitionRatio += partition_ratio;
                total_partitionSize += ts.tuple_size_total;
                Column catalog_col = ReplicatedColumn.get(catalog_tbl);
                pentry = new TableEntry(PartitionMethodType.REPLICATION, catalog_col);
                if (debug.val)
                    LOG.debug(String.format("Replicating %s at all partitions [%s]", catalog_tbl.getName(), catalog_col.fullName()));

                // -------------------------------
                // Forced Selection
                // -------------------------------
            } else if (forced_columns.isEmpty() == false) {
                // Assume there is only one candidate
                assert (forced_columns.size() == 1) : "Unexpected number of forced columns: " + forced_columns;
                Column catalog_col = CollectionUtil.first(forced_columns);
                pentry = new TableEntry(PartitionMethodType.HASH, catalog_col);
                total_partitionRatio += partition_size / (double) hints.max_memory_per_partition;
                total_partitionSize += partition_size;
                if (debug.val)
                    LOG.debug(String.format("Forcing %s to be partitioned by specific column [%s]", catalog_tbl.getName(), catalog_col.fullName()));

                // -------------------------------
                // Select Most Popular
                // -------------------------------
            } else {
                // If there are no edges, then we'll just randomly pick a column
                // since it doesn't matter
                final Collection<DesignerEdge> edges = agraph.getIncidentEdges(v);
                if (edges.isEmpty())
                    continue;
                if (trace.val)
                    LOG.trace(catalog_tbl + " has " + edges.size() + " edges in AccessGraph");

                ObjectHistogram<Column> column_histogram = null;
                ObjectHistogram<Column> join_column_histogram = new ObjectHistogram<Column>();
                ObjectHistogram<Column> self_column_histogram = new ObjectHistogram<Column>();
                // Map<Column, Double> unsorted = new HashMap<Column, Double>();
                for (DesignerEdge e : edges) {
                    Collection<DesignerVertex> vertices = agraph.getIncidentVertices(e);
                    DesignerVertex v0 = CollectionUtil.get(vertices, 0);
                    DesignerVertex v1 = CollectionUtil.get(vertices, 1);
                    boolean self = (v0.equals(v) && v1.equals(v));
                    column_histogram = (self ? self_column_histogram : join_column_histogram);

                    double edge_weight = e.getTotalWeight();
                    PredicatePairs cset = e.getAttribute(AccessGraph.EdgeAttributes.COLUMNSET);
                    if (trace.val)
                        LOG.trace("Examining ColumnSet for " + e.toString(true));

                    Histogram<Column> cset_histogram = cset.buildHistogramForType(Column.class);
                    Collection<Column> columns = cset_histogram.values();
                    if (trace.val)
                        LOG.trace("Constructed Histogram for " + catalog_tbl + " from ColumnSet:\n"
                                + cset_histogram.setDebugLabels(CatalogUtil.getHistogramLabels(cset_histogram.values())).toString(100, 50));
                    for (Column catalog_col : columns) {
                        if (!catalog_col.getParent().equals(catalog_tbl))
                            continue;
                        if (catalog_col.getNullable())
                            continue;
                        long cnt = cset_histogram.get(catalog_col);
                        if (trace.val)
                            LOG.trace("Found Match: " + catalog_col.fullName() + " [cnt=" + cnt + "]");
                        column_histogram.put(catalog_col, Math.round(cnt * edge_weight));
                    } // FOR
                    // System.err.println(cset.debug());
                    // LOG.info("[" + e.getTotalWeight() + "]: " + cset);
                } // FOR

                // If there were no join columns, then use the self-reference
                // histogram
                column_histogram = (join_column_histogram.isEmpty() ? self_column_histogram : join_column_histogram);
                if (column_histogram.isEmpty()) {
                    EventObserver<DesignerVertex> observer = new EventObserver<DesignerVertex>() {
                        @Override
                        public void update(EventObservable<DesignerVertex> o, DesignerVertex v) {
                            for (DesignerEdge e : agraph.getIncidentEdges(v)) {
                                LOG.info(e.getAttribute(AccessGraph.EdgeAttributes.COLUMNSET));
                            }
                            LOG.info(StringUtil.repeat("-", 100));
                        }
                    };
                    LOG.info("Edges: " + edges);
                    GraphVisualizationPanel.createFrame(agraph, observer).setVisible(true);
                    // ThreadUtil.sleep(10000);
                }

                // We might not find anything if we are calculating the lower
                // bounds using only one transaction
                // if (column_histogram.isEmpty()) {
                // if (trace.val)
                // LOG.trace("Failed to find any ColumnSets for " +
                // catalog_tbl);
                // continue;
                // }
                assert (!column_histogram.isEmpty()) : "Failed to find any ColumnSets for " + catalog_tbl;
                if (trace.val)
                    LOG.trace("Column Histogram:\n" + column_histogram);

                Column catalog_col = CollectionUtil.first(column_histogram.getMaxCountValues());
                pentry = new TableEntry(PartitionMethodType.HASH, catalog_col, null, null);
                total_partitionRatio += partition_size / (double) hints.max_memory_per_partition;
                total_partitionSize += partition_size;

                if (debug.val)
                    LOG.debug(String.format("Selected %s's most popular column for partitioning [%s]", catalog_tbl.getName(), catalog_col.fullName()));
            }
            pplan.table_entries.put(catalog_tbl, pentry);

            if (debug.val)
                LOG.debug(String.format("Current Partition Size: %s", StringUtil.formatSize(total_partitionSize), StringUtil.formatSize(hints.max_memory_per_partition)));
            assert (total_partitionRatio <= 1) : String.format("Too much memory per partition: %s / %s", StringUtil.formatSize(total_partitionSize),
                    StringUtil.formatSize(hints.max_memory_per_partition));
        } // FOR

        for (Table catalog_tbl : info.catalogContext.database.getTables()) {
            if (pplan.getTableEntry(catalog_tbl) == null) {
                Column catalog_col = CollectionUtil.random(catalog_tbl.getColumns());
                assert (catalog_col != null) : "Failed to randomly pick column for " + catalog_tbl;
                pplan.table_entries.put(catalog_tbl, new TableEntry(PartitionMethodType.HASH, catalog_col, null, null));
                if (debug.val)
                    LOG.debug(String.format("No partitioning column selected for %s. Choosing a random attribute [%s]", catalog_tbl, catalog_col.fullName()));
            }
        } // FOR

View Full Code Here
TOP

Related Classes of edu.brown.designer.partitioners.plan.TableEntry

  • edu.brown.designer.generators.PartitionPlanTreeGenerator
  • edu.brown.designer.partitioners.HeuristicPartitioner
  • edu.brown.designer.partitioners.MostPopularPartitioner
  • edu.brown.designer.partitioners.PrimaryKeyPartitioner
  • edu.brown.designer.partitioners.TestPrimaryKeyPartitioner
  • edu.brown.gui.designer.PartitionPlanPanel
    • Copyright © 2018 www.massapicom. All rights reserved.
    All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.