From 3800973a69c23deb5c770e8032074244f50f5b9a Mon Sep 17 00:00:00 2001
From: Valery Kharseko <vharseko@3a-systems.ru>
Date: Thu, 20 Aug 2026 09:02:47 +0000
Subject: [PATCH] Stamp JDBC backend tables with their tree name and refresh optimizer statistics after import (#866)
---
opendj-server-legacy/src/main/java/org/opends/server/backends/jdbc/JDBCStorage.java | 645 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 636 insertions(+), 9 deletions(-)
diff --git a/opendj-server-legacy/src/main/java/org/opends/server/backends/jdbc/JDBCStorage.java b/opendj-server-legacy/src/main/java/org/opends/server/backends/jdbc/JDBCStorage.java
index 0987886..f0e6557 100644
--- a/opendj-server-legacy/src/main/java/org/opends/server/backends/jdbc/JDBCStorage.java
+++ b/opendj-server-legacy/src/main/java/org/opends/server/backends/jdbc/JDBCStorage.java
@@ -34,11 +34,13 @@
import org.opends.server.types.RestoreConfig;
import org.opends.server.util.BackupManager;
+import java.io.Closeable;
import java.nio.ByteBuffer;
import java.security.MessageDigest;
import java.security.NoSuchAlgorithmException;
import java.sql.*;
import java.util.*;
+import java.util.concurrent.ConcurrentHashMap;
import static org.opends.server.backends.pluggable.spi.StorageUtils.addErrorMessage;
import static org.opends.server.util.StaticUtils.stackTraceToSingleLineString;
@@ -88,6 +90,14 @@
return statement.executeUpdate();
}
+ // unlike execute(), tolerates statements that return a result set ("analyze table" on mysql)
+ void executeAny(PreparedStatement statement) throws SQLException {
+ if (logger.isTraceEnabled()) {
+ logger.trace(LocalizableMessage.raw("jdbc: %s",statement));
+ }
+ statement.execute();
+ }
+
Connection getConnection() throws Exception {
return CachedConnection.getConnection(config.getDBDirectory());
}
@@ -111,6 +121,10 @@
@Override
public void close() {
storageStatus = StorageStatus.lockedDown(LocalizableMessage.raw("closed"));
+ // a stamp that the database rejected is remembered for as long as the storage is open, so
+ // that it is not reissued for every tree on every open; disabling and re-enabling the
+ // backend is the way to try again once the privilege has been granted
+ unstampableTrees.clear();
}
final LoadingCache<TreeName,String> tree2table = Caffeine.newBuilder()
@@ -151,6 +165,567 @@
return name;
}
+ private static final String[] NO_ARGS=new String[0];
+
+ // Comment statements take a lock (a metadata lock on mysql, a schema modification lock on sql
+ // server, a ddl lock on oracle), and mysql and sql server wait for it without limit by
+ // default (lock_wait_timeout is a year, lock_timeout is infinite): a stamp could queue behind
+ // an unrelated transaction of another session on the same database and - on mysql - park
+ // every other query on the table behind itself. The stamp is a diagnostic aid, so every
+ // dialect is told to give up after this many seconds instead of waiting.
+ private static final int COMMENT_LOCK_TIMEOUT_SECONDS=5;
+
+ // The comment statement runs on a connection of its own (newStampConnection() below), and a
+ // driver waits for a connect attempt without limit unless it is told otherwise: a database
+ // that keeps its established connections alive but accepts no new ones (a moved vip, a proxy
+ // at its connection limit) would otherwise hang the open of a tree - dsconfig
+ // create-backend-index opens one on a running server - instead of leaving a table unstamped.
+ // Every dialect gets the same bound, in the unit its own driver property takes.
+ private static final int STAMP_CONNECT_TIMEOUT_SECONDS=10;
+
+ // Not one of the four drivers bounds the whole login attempt with its connect property alone:
+ // postgres and mysql apply theirs to socket.connect(), oracle's own reference says
+ // CONNECT_TIMEOUT "doesn't include user authentication", and the sql server driver leaves the
+ // read of the prelogin answer unbounded - TDSChannel.open() gives the socket
+ // min(what is left of loginTimeout, socketTimeout) and socketTimeout defaults to 0, which is
+ // "wait forever". Those reads - of the prelogin handshake, of tls, of authentication - are
+ // exactly where a proxy that accepts a connection and then goes quiet leaves the driver, so
+ // every dialect carries a read bound as well. All four are socket read timeouts, so the bound
+ // outlives the login phase and covers the comment statement too, which is why it is kept well
+ // clear of the lock bound above: the statement gives up on a contended lock long before the
+ // socket gives up on the server. On a mysql connection with tls (the sslMode=PREFERRED default
+ // of connector/j) the wall clock of a dead peer is twice this, since closing an SSLSocket
+ // drains input waiting for close_notify and pays the read bound a second time.
+ private static final int STAMP_READ_TIMEOUT_SECONDS=30;
+
+ // Trees whose stamp failed for a reason that is not going to change by itself: an account that
+ // may not comment its tables (no ALTER privilege, for instance) would otherwise reissue the
+ // statement for every tree on every open. A failure that says nothing about the table - a lock
+ // timeout, a connection that broke - is not remembered (failureScope() below), so a contended
+ // moment does not leave the backend unstamped until it is restarted. Forgotten when the
+ // storage is closed, so re-enabling the backend is enough to try again once the privilege has
+ // been granted, without a restart of the server.
+ private final Set<TreeName> unstampableTrees=ConcurrentHashMap.newKeySet();
+
+ /**
+ * The engines whose comment statement, comment readback and statistics refresh this backend
+ * knows, with the session settings a comment statement needs: the driver properties that
+ * bound the connect attempt of the connection it runs on, and the statement that bounds its
+ * wait for the table lock.
+ */
+ enum Dialect {
+ /** postgresql: lock_timeout takes milliseconds; connectTimeout bounds socket.connect(), loginTimeout the whole login the driver runs on a thread of its own, socketTimeout every read after it - all three in seconds. */
+ POSTGRES("set lock_timeout = "+(COMMENT_LOCK_TIMEOUT_SECONDS*1000),
+ "connectTimeout", STAMP_CONNECT_TIMEOUT_SECONDS, "loginTimeout", STAMP_CONNECT_TIMEOUT_SECONDS,
+ "socketTimeout", STAMP_READ_TIMEOUT_SECONDS),
+ /** mysql: lock_wait_timeout takes seconds; connectTimeout bounds the socket connect and socketTimeout every read after it, both in milliseconds. */
+ MYSQL("set session lock_wait_timeout="+COMMENT_LOCK_TIMEOUT_SECONDS,
+ "connectTimeout", STAMP_CONNECT_TIMEOUT_SECONDS*1000, "socketTimeout", STAMP_READ_TIMEOUT_SECONDS*1000),
+ /** oracle: ddl_lock_timeout takes seconds and defaults to 0 (give up at once), but it can be raised globally; the connect and read bounds take milliseconds. */
+ ORACLE("alter session set ddl_lock_timeout="+COMMENT_LOCK_TIMEOUT_SECONDS,
+ "oracle.net.CONNECT_TIMEOUT", STAMP_CONNECT_TIMEOUT_SECONDS*1000, "oracle.jdbc.ReadTimeout", STAMP_READ_TIMEOUT_SECONDS*1000),
+ /** ms sql server: lock_timeout takes milliseconds; loginTimeout bounds the socket connect, in seconds, and socketTimeout the prelogin read it leaves open - and every read after it - in milliseconds. */
+ MICROSOFT("set lock_timeout "+(COMMENT_LOCK_TIMEOUT_SECONDS*1000),
+ "loginTimeout", STAMP_CONNECT_TIMEOUT_SECONDS, "socketTimeout", STAMP_READ_TIMEOUT_SECONDS*1000);
+
+ final String lockTimeoutSql;
+ // The driver properties bounding the login attempt of a stamp connection: the one bounding
+ // the socket connect and the one bounding the reads behind it, each in the unit its own
+ // driver takes. newStampConnection() hands the driver a copy of them: a driver is free to
+ // write into the map it is passed, and the sql server one gives a supplied property
+ // precedence over the same property of the url.
+ final Properties connectProperties=new Properties();
+
+ /** For the three drivers whose connect property and read property bound the login between them. */
+ Dialect(String lockTimeoutSql, String connectProperty, int connectValue, String readProperty, int readValue) {
+ this.lockTimeoutSql=lockTimeoutSql;
+ connectProperties.setProperty(connectProperty, String.valueOf(connectValue));
+ connectProperties.setProperty(readProperty, String.valueOf(readValue));
+ }
+
+ /** For the one driver bounding the login itself, on top of the socket connect and the reads behind it. */
+ Dialect(String lockTimeoutSql, String connectProperty, int connectValue, String loginProperty, int loginValue,
+ String readProperty, int readValue) {
+ this(lockTimeoutSql, connectProperty, connectValue, readProperty, readValue);
+ connectProperties.setProperty(loginProperty, String.valueOf(loginValue));
+ }
+ }
+
+ static String driverNameOf(Connection con) {
+ return ((CachedConnection) con).parent.getClass().getName();
+ }
+
+ // The dialect behind a pooled connection, or null for an engine none of the statements of this
+ // class fit: it is left unstamped and its statistics untouched rather than fed untested SQL.
+ static Dialect dialectOf(Connection con) {
+ final String driverName=driverNameOf(con);
+ if (driverName.contains("postgres")) {
+ return Dialect.POSTGRES;
+ }else if (driverName.contains("mysql")) {
+ return Dialect.MYSQL;
+ }else if (driverName.contains("oracle")) {
+ return Dialect.ORACLE;
+ }else if (driverName.contains("microsoft")) {
+ return Dialect.MICROSOFT;
+ }
+ return null;
+ }
+
+ /** Outcome of a comment stamp: openTree() ignores it, tests tell the cases apart. */
+ enum CommentResult {
+ /** the table now carries its tree name */
+ STAMPED,
+ /** the stored comment already matched: no statement was issued */
+ UP_TO_DATE,
+ /** neither comment syntax nor readback is known for this engine */
+ UNSUPPORTED,
+ /** the comment could not be read back or stored */
+ FAILED
+ }
+
+ // Splices a value into a single-quoted SQL literal for the comment DDL, which takes no bind
+ // parameters: doubles every quote, and every backslash on dialects where backslash is an
+ // escape character inside literals. The scan of the escaped result is defence in depth: it
+ // re-verifies that no quote (or live backslash) is left unpaired and able to terminate the
+ // literal, so a regression in the escaping throws instead of reaching the database.
+ private static String sqlLiteral(String value, boolean backslashIsEscape) {
+ final String escaped=(backslashIsEscape?value.replace("\\","\\\\"):value).replace("'","''");
+ for (int i=0;i<escaped.length();i++) {
+ final char c=escaped.charAt(i);
+ if (c=='\'' || (backslashIsEscape && c=='\\')) {
+ if (i+1>=escaped.length() || escaped.charAt(i+1)!=c) {
+ throw new IllegalArgumentException("unpaired "+c+" in SQL literal: "+escaped);
+ }
+ i++;
+ }
+ }
+ return "'"+escaped+"'";
+ }
+
+ // Whether backslash is an escape character inside string literals on this mysql connection:
+ // under the NO_BACKSLASH_ESCAPES sql mode it is an ordinary character, and doubling it there
+ // would store a comment that never matches its tree name - re-stamping the table forever.
+ // Asked of the very session that parses the literal: sql_mode is a session setting, and a
+ // session opened at another moment can have been given another value of it.
+ boolean isMysqlBackslashEscape(Connection con) throws SQLException {
+ try (final PreparedStatement statement=con.prepareStatement("select @@sql_mode");
+ final ResultSet rs=executeResultSet(statement)) {
+ final String sqlMode=rs.next() ? rs.getString(1) : null;
+ return sqlMode==null || !sqlMode.toUpperCase().contains("NO_BACKSLASH_ESCAPES");
+ }
+ }
+
+ // A connection of its own for the comment statements, outside the pool: they need session
+ // settings (the lock timeout above) that a pooled connection would carry over to whoever
+ // borrows it next, since CachedConnection.close() only rolls back.
+ Connection newStampConnection(Dialect dialect) throws SQLException {
+ final Properties properties=new Properties();
+ properties.putAll(dialect.connectProperties);
+ final Connection con=DriverManager.getConnection(config.getDBDirectory(), properties);
+ try {
+ con.setAutoCommit(false);
+ executeSessionStatement(con, dialect.lockTimeoutSql); // give up instead of waiting for another session
+ // postgres undoes a plain SET when the transaction that ran it is rolled back, and a
+ // failed stamp is rolled back with the connection kept (StampSession.reset() below):
+ // commit the setting, or the first failure of a sweep would leave every tree after it
+ // stamped without the very bound this connection exists to carry. The session settings
+ // of the other three dialects are not transactional - the commit costs them an empty
+ // transaction.
+ con.commit();
+ }catch (SQLException e) { // nothing else holds this connection yet: it would leak
+ try {
+ con.close();
+ }catch (SQLException e2) {}
+ throw e;
+ }
+ return con;
+ }
+
+ // The connection the comment statements of one sweep of openTree() calls share. Opening a
+ // backend opens every tree it holds (about 25 for a stock suffix), so a connection per stamp
+ // would mean that many physical connects on the first open after an upgrade - the one open
+ // that stamps them all. Opened lazily: a sweep that finds every comment up to date, which is
+ // every open after the first, opens nothing at all.
+ final class StampSession implements Closeable {
+ private Connection con;
+
+ // Whether backslash escapes inside a literal on the connection above (mysql @@sql_mode).
+ // It is a session setting of a connection the whole sweep shares, so the sweep asks once
+ // instead of once per tree, and forgets it together with the session it describes.
+ private Boolean mysqlBackslashEscape;
+
+ // Set when a stamp failed for a reason no other tree of this sweep would escape either: a
+ // connect that did not go through, a lock the statement gave up on. Each remaining tree
+ // would pay that same bound - or that same connect attempt - again, which is a backend
+ // open held for the bound times the number of its trees, all for a diagnostic aid. The
+ // sweep gives up instead; nothing about the trees is remembered, so the next open retries.
+ private boolean gaveUp;
+
+ Connection connection(Dialect dialect) throws SQLException {
+ if (con==null) {
+ con=newStampConnection(dialect);
+ }
+ return con;
+ }
+
+ // Asked by the mysql statement only, and only once the connection above is open.
+ boolean backslashIsEscape() throws SQLException {
+ if (mysqlBackslashEscape==null) {
+ mysqlBackslashEscape=isMysqlBackslashEscape(con);
+ }
+ return mysqlBackslashEscape;
+ }
+
+ void giveUp() {
+ gaveUp=true;
+ }
+
+ boolean hasGivenUp() {
+ return gaveUp;
+ }
+
+ // A statement that failed can leave the session unusable (postgres refuses every further
+ // statement of the transaction with 25P02 until it is rolled back), so the stamp of the
+ // next tree gets a clean one: rolled back, or replaced when even the rollback fails.
+ void reset() {
+ if (con!=null) {
+ try {
+ con.rollback();
+ }catch (SQLException e) {
+ close();
+ }
+ }
+ }
+
+ @Override
+ public void close() {
+ if (con!=null) {
+ try {
+ con.close();
+ }catch (SQLException e) {
+ logger.trace(LocalizableMessage.raw("jdbc: unable to close the comment connection: %s", stackTraceToSingleLineString(e)));
+ }
+ con=null;
+ }
+ mysqlBackslashEscape=null; // it described the session that has just gone
+ }
+ }
+
+ // A session setting must reach the server as a plain batch: the sql server driver runs a
+ // prepared statement through sp_executesql, and a setting made there is reverted when that
+ // call returns - before the statement it is meant to protect ever runs.
+ private void executeSessionStatement(Connection con, String sql) throws SQLException {
+ try (final Statement statement=con.createStatement()) {
+ if (logger.isTraceEnabled()) {
+ logger.trace(LocalizableMessage.raw("jdbc: %s",sql));
+ }
+ statement.execute(sql);
+ }
+ }
+
+ // Table names are opaque SHA-224 hashes, so on the database side there is no way to tell
+ // which tree a table holds. Stamp each table with its tree name (visible in "\dt+" and the
+ // information schema) so database-level troubleshooting does not require recomputing hashes.
+ // Runs on a dedicated connection, never on the transaction that opened the tree: comment
+ // statements are DDL (an implicit commit on mysql and oracle), and a failing
+ // sp_addextendedproperty rolls the whole transaction back on sql server - either would
+ // corrupt work pending on the caller's connection (e.g. the trusted flag written by
+ // DefaultIndex.afterOpen()). The comment is a diagnostic aid: a failed attempt only logs and
+ // must not fail the backend.
+ CommentResult commentTable(TreeName treeName, Dialect dialect) {
+ try (final StampSession session=new StampSession()) { // a stamp of its own: no sweep to share a connection with
+ return commentTable(treeName, dialect, session);
+ }
+ }
+
+ CommentResult commentTable(TreeName treeName, Dialect dialect, StampSession session) {
+ final String tableName=getTableName(treeName);
+ if (dialect==null) { // no comment syntax and readback known for other engines: leave the table unstamped
+ return CommentResult.UNSUPPORTED;
+ }
+ if (unstampableTrees.contains(treeName)) { // the database already rejected this one: do not ask again
+ return CommentResult.FAILED;
+ }
+ if (session.hasGivenUp()) { // an earlier tree of this sweep lost the session every tree of it needs
+ logger.debug(LocalizableMessage.raw("jdbc: table %s is left unstamped: the stamp of an earlier table of this open lost its connection", tableName));
+ return CommentResult.FAILED;
+ }
+ final String treeComment=treeName.toString();
+ try {
+ // The readback runs on the stamp connection, not on one borrowed from the pool: the
+ // caller of openTree() is inside a transaction and holding a pooled connection already,
+ // and a pool that cannot open a second one waits for a peer to return one - which here
+ // is the very thread that is waiting. The dialect comes from the caller's connection
+ // for the same reason: finding it out must not cost a borrow either.
+ final Connection con=session.connection(dialect);
+ // comment statements are DDL (metadata lock on mysql, ddl lock on oracle) and openTree()
+ // runs on every backend open: only stamp when the stored comment is absent or stale
+ final String storedComment=readStoredComment(con, dialect, tableName);
+ // end the read: this connection is shared by every tree of the sweep and must not hold
+ // a transaction open across all of them
+ con.commit();
+ if (treeComment.equals(storedComment)) {
+ return CommentResult.UP_TO_DATE;
+ }
+ final String sql;
+ final String[] args;
+ switch (dialect) {
+ case MYSQL: // ALTER TABLE takes no binds; whether backslash escapes inside the literal depends on the sql mode of this session
+ sql="alter table "+tableName+" comment "+sqlLiteral(treeComment,session.backslashIsEscape());
+ args=NO_ARGS;
+ break;
+ case MICROSOFT: // no COMMENT ON in t-sql: MS_Description extended property (procedure arguments take binds)
+ sql="declare @s sysname = schema_name()"
+ +" if exists (select 1 from sys.extended_properties where class=1 and major_id=object_id(?) and minor_id=0 and name='MS_Description')"
+ +" exec sys.sp_updateextendedproperty N'MS_Description', ?, N'SCHEMA', @s, N'TABLE', ?"
+ +" else"
+ +" exec sys.sp_addextendedproperty N'MS_Description', ?, N'SCHEMA', @s, N'TABLE', ?";
+ args=new String[]{tableName, treeComment, tableName, treeComment, tableName};
+ break;
+ case POSTGRES: // no binds in ddl; the E'' form keeps backslash an escape character regardless of standard_conforming_strings
+ sql="comment on table "+tableName+" is E"+sqlLiteral(treeComment,true);
+ args=NO_ARGS;
+ break;
+ case ORACLE: // no binds in ddl; backslash is never an escape character in oracle literals
+ sql="comment on table "+tableName+" is "+sqlLiteral(treeComment,false);
+ args=NO_ARGS;
+ break;
+ default: // a dialect this switch was never told about must not inherit another one's ddl
+ throw new IllegalStateException("no comment statement for dialect "+dialect);
+ }
+ try (final PreparedStatement statement=con.prepareStatement(sql)) {
+ for (int i=0;i<args.length;i++) {
+ statement.setString(i+1,args[i]);
+ }
+ executeAny(statement);
+ con.commit();
+ }
+ return CommentResult.STAMPED;
+ }catch (Exception e) {
+ session.reset(); // what the failed statement left behind must not poison the stamp of the next tree
+ final FailureScope scope=failureScope(e, dialect);
+ if (scope==FailureScope.SESSION) {
+ // the connection itself is gone, and every tree left in this sweep needs one: each
+ // would pay the same connect attempt again, ~25 of them for a stock suffix
+ session.giveUp();
+ }else if (scope==FailureScope.TREE) {
+ unstampableTrees.add(treeName);
+ }
+ logger.warn(LocalizableMessage.raw("jdbc: unable to comment table %s with tree name %s, it stays unstamped %s (the comment is a diagnostic aid: the backend is unaffected): %s",
+ tableName, treeName, scope==FailureScope.TREE?"until this backend is closed":"for now", stackTraceToSingleLineString(e)));
+ return CommentResult.FAILED;
+ }
+ }
+
+ /** What a failed stamp says about stamping again - this tree, and the trees behind it in the same sweep. */
+ enum FailureScope {
+ /**
+ * The database rejected the statement: an account that may not comment its tables, say. It
+ * would be rejected again for this tree on every open, so the tree is remembered and not
+ * asked again while this backend is open. Says nothing about the other trees of the sweep,
+ * which are stamped as usual - the privilege may well be missing for this one table alone.
+ */
+ TREE,
+ /**
+ * Another session held the table locked and the stamp gave up on the bound above. Nothing
+ * is remembered - the next open tries again - and the sweep goes on: the lock belongs to
+ * this table, and the trees behind it are no more likely to be contended than usual.
+ */
+ MOMENT,
+ /**
+ * The connection the sweep runs on is gone, or was never established. Every tree left in
+ * the sweep would run into the same thing, one connect attempt each, so the sweep ends;
+ * nothing is remembered, since this says nothing about any of the tables.
+ */
+ SESSION
+ }
+
+ // What a failed stamp says about trying again. Both chains of the failure are walked: a driver
+ // reports the vendor error of a rejected statement as the next exception of a generic one at
+ // least as often as it reports it as the cause, and reading only one of the two would classify
+ // a lock timeout as a rejection, which leaves the tree unstamped for the life of the backend
+ // over a moment of contention.
+ static FailureScope failureScope(Throwable failure, Dialect dialect) {
+ FailureScope scope=FailureScope.TREE;
+ final Deque<Throwable> pending=new ArrayDeque<>();
+ final Set<Throwable> seen=Collections.newSetFromMap(new IdentityHashMap<Throwable,Boolean>());
+ if (failure!=null) {
+ pending.push(failure);
+ }
+ while (!pending.isEmpty()) {
+ final Throwable e=pending.pop();
+ if (!seen.add(e)) { // a driver that chains an exception back to itself must not loop this walk
+ continue;
+ }
+ if (e.getCause()!=null) {
+ pending.push(e.getCause());
+ }
+ if (!(e instanceof SQLException)) {
+ continue;
+ }
+ final SQLException sqlException=(SQLException) e;
+ if (sqlException.getNextException()!=null) {
+ pending.push(sqlException.getNextException());
+ }
+ final FailureScope found=scopeOf(sqlException, dialect);
+ if (found==FailureScope.SESSION) { // nothing further down either chain can weaken this one
+ return FailureScope.SESSION;
+ }
+ if (found==FailureScope.MOMENT) {
+ scope=FailureScope.MOMENT;
+ }
+ }
+ return scope;
+ }
+
+ // What one exception of the chain says on its own.
+ private static FailureScope scopeOf(SQLException e, Dialect dialect) {
+ final String sqlState=e.getSQLState();
+ if (e instanceof SQLTransientConnectionException || e instanceof SQLNonTransientConnectionException
+ || e instanceof SQLRecoverableException // what oracle throws for a connection that has gone
+ || (sqlState!=null && sqlState.startsWith("08"))) { // connection exception
+ return FailureScope.SESSION;
+ }
+ if (e instanceof SQLTimeoutException || e instanceof SQLTransientException) {
+ return FailureScope.MOMENT;
+ }
+ if (dialect==null) { // the failure came before the engine was known
+ return FailureScope.TREE;
+ }
+ switch (dialect) {
+ case POSTGRES: // 55P03 lock not available: lock_timeout expired
+ return "55P03".equals(sqlState) ? FailureScope.MOMENT : FailureScope.TREE;
+ case MYSQL: // 1205 lock wait timeout exceeded
+ return e.getErrorCode()==1205 ? FailureScope.MOMENT : FailureScope.TREE;
+ case ORACLE: // ORA-00054 resource busy, ORA-04021 timeout occurred while waiting to lock object
+ return e.getErrorCode()==54 || e.getErrorCode()==4021 ? FailureScope.MOMENT : FailureScope.TREE;
+ case MICROSOFT: // 1222 lock request time out period exceeded
+ return e.getErrorCode()==1222 ? FailureScope.MOMENT : FailureScope.TREE;
+ default: // a dialect with no lock timeout code of its own here: its failures are not treated as ones of the moment
+ return FailureScope.TREE;
+ }
+ }
+
+ // Returns the comment currently stored on the table, or null when there is none. The dialect is
+ // passed in rather than read off the connection: this runs on the stamp connection, which is
+ // not a pooled one, and only for the dialects commentTable() recognizes.
+ String readStoredComment(Connection con, Dialect dialect, String tableName) throws SQLException {
+ final String sql;
+ final String arg;
+ switch (dialect) {
+ case POSTGRES:
+ sql="select obj_description(to_regclass(?), 'pg_class')";
+ arg=tableName;
+ break;
+ case MYSQL:
+ sql="select table_comment from information_schema.tables where table_schema=database() and table_name=?";
+ arg=tableName;
+ break;
+ case ORACLE:
+ sql="select comments from user_tab_comments where table_name=?";
+ arg=tableName.toUpperCase();
+ break;
+ case MICROSOFT:
+ sql="select cast(value as nvarchar(4000)) from sys.extended_properties where class=1 and major_id=object_id(?) and minor_id=0 and name='MS_Description'";
+ arg=tableName;
+ break;
+ default: // a dialect this switch was never told about must not inherit another one's catalog
+ throw new IllegalStateException("no table comment readback for dialect "+dialect);
+ }
+ try (final PreparedStatement statement=con.prepareStatement(sql)) {
+ statement.setString(1,arg);
+ try (final ResultSet rs=executeResultSet(statement)) {
+ return rs.next() ? rs.getString(1) : null;
+ }
+ }
+ }
+
+ // Statistics upkeep after an import is bounded and can be turned off: gathering statistics of
+ // a freshly loaded table is a full scan on oracle (dbms_stats defaults to AUTO_SAMPLE_SIZE,
+ // and the entries themselves live in the blob column it reads), which a multi-million entry
+ // backend would otherwise pay in full, with no way to cap or skip it, after import-ldif has
+ // already reported its final status.
+ static final String STATISTICS_PROPERTY="org.openidentityplatform.opendj.jdbc.statistics";
+ static final String STATISTICS_TIMEOUT_PROPERTY=STATISTICS_PROPERTY+".timeout";
+ private static final int STATISTICS_TIMEOUT_SECONDS_DEFAULT=600;
+
+ // A bulk load leaves the optimizer statistics of freshly created tables stale (a table that
+ // was never analyzed can make the planner badly misestimate the "where k>? order by k" cursor
+ // batches - see OpenIdentityPlatform/OpenDJ#859), so refresh them once the data is in place.
+ // Only the trees the import actually wrote are refreshed: rebuild-index imports a few index
+ // trees, and gathering statistics of the whole backend on its behalf is a full scan per
+ // table on oracle. Statistics upkeep is best-effort: a failure must not fail the import that
+ // produced the data, so failures are only logged - the return value makes them observable to tests.
+ boolean updateTableStatistics(Connection con, Collection<TreeName> trees) {
+ if (!Boolean.parseBoolean(System.getProperty(STATISTICS_PROPERTY,"true"))) {
+ logger.debug(LocalizableMessage.raw("jdbc: statistics refresh turned off by %s", STATISTICS_PROPERTY));
+ return false; // nothing was refreshed
+ }
+ final Dialect dialect=dialectOf(con);
+ if (dialect==null) { // no portable statistics refresh for other engines
+ return false; // nothing was refreshed: reporting success here would make the assertion of the tests vacuous
+ }
+ final int timeoutSeconds=Math.max(0,Integer.getInteger(STATISTICS_TIMEOUT_PROPERTY,STATISTICS_TIMEOUT_SECONDS_DEFAULT));
+ boolean allRefreshed=true;
+ for (final TreeName treeName : trees) {
+ final String tableName=getTableName(treeName);
+ // The statement is chosen inside the try, so that the guard of the default branch
+ // degrades to "this table was not refreshed" like every other failure here: the
+ // contract above is that a refresh which failed never fails the import that produced
+ // the data, and a throw escaping this loop would break it.
+ try {
+ final String sql;
+ final String[] args;
+ switch (dialect) {
+ case POSTGRES:
+ sql="analyze "+tableName;
+ args=NO_ARGS;
+ break;
+ case MYSQL:
+ sql="analyze table "+tableName;
+ args=NO_ARGS;
+ break;
+ case ORACLE:
+ sql="begin dbms_stats.gather_table_stats(user, ?); end;";
+ args=new String[]{tableName.toUpperCase()};
+ break;
+ case MICROSOFT:
+ sql="update statistics "+tableName;
+ args=NO_ARGS;
+ break;
+ default: // a dialect this switch was never told about must not inherit another one's statement
+ throw new IllegalStateException("no statistics refresh for dialect "+dialect);
+ }
+ try (final PreparedStatement statement=con.prepareStatement(sql)) {
+ statement.setQueryTimeout(timeoutSeconds); // 0: wait without limit
+ for (int i=0;i<args.length;i++) {
+ statement.setString(i+1,args[i]);
+ }
+ if (dialect==Dialect.MYSQL) { // mysql reports analyze problems as a result row, not an SQLException
+ try (final ResultSet rs=executeResultSet(statement)) {
+ while (rs.next()) {
+ if ("error".equalsIgnoreCase(rs.getString("Msg_type"))) {
+ throw new SQLException(rs.getString("Msg_text"));
+ }
+ }
+ }
+ }else {
+ executeAny(statement);
+ }
+ con.commit();
+ }
+ }catch (Exception e) {
+ try {
+ con.rollback();
+ } catch (SQLException e2) {}
+ allRefreshed=false;
+ logger.warn(LocalizableMessage.raw("jdbc: unable to refresh statistics of table %s (tree %s): %s",
+ tableName, treeName, stackTraceToSingleLineString(e)));
+ }
+ }
+ return allRefreshed;
+ }
+
@Override
public void removeStorageFiles() throws StorageRuntimeException {
final boolean isOpen=getStorageStatus().isWorking();
@@ -180,6 +755,11 @@
} catch (Exception e) {
throw new StorageRuntimeException(e);
}
+ // all tables are gone: forget the mappings so listTrees() consumers skip the dropped trees
+ for (final TreeName treeName : trees) {
+ tree2table.invalidate(treeName);
+ unstampableTrees.remove(treeName); // a table recreated later deserves a fresh stamp attempt
+ }
}
if (!isOpen) {
close();
@@ -197,14 +777,17 @@
@Override
public void write(WriteOperation writeOperation) throws Exception {
try (final Connection con=getConnection()) {
+ final WriteableTransactionTransactionImpl txn=new WriteableTransactionTransactionImpl(con);
try {
- writeOperation.run(new WriteableTransactionTransactionImpl(con));
+ writeOperation.run(txn);
con.commit();
} catch (Exception e) {
try {
con.rollback();
} catch (SQLException ex) {}
throw e;
+ } finally { // the comment connection lives no longer than the trees it stamped
+ txn.stampSession.close();
}
}
}
@@ -274,6 +857,11 @@
}
private final class WriteableTransactionTransactionImpl extends ReadableTransactionImpl implements WriteableTransaction {
+ // Shared by every table this transaction stamps: opening a backend opens all its trees,
+ // and each stamp of its own connection would be a physical connect of its own. Closed by
+ // write() (and by ImporterImpl.close()) when the transaction is done with.
+ final StampSession stampSession=new StampSession();
+
public WriteableTransactionTransactionImpl(Connection con) {
super(con);
if (!accessMode.isWriteable()) {
@@ -363,6 +951,9 @@
}
}
// mssql: k is varbinary(max), which cannot be an index key column - cursor batches stay unindexed there
+ // the dialect is taken off this transaction's own connection: finding it out must
+ // not cost a borrow from a pool this thread is already holding a connection of
+ commentTable(treeName, dialectOf(con), stampSession);
}
}
@@ -397,6 +988,9 @@
throw new StorageRuntimeException(e);
}
}
+ // forget the mapping so listTrees() consumers (updateTableStatistics) skip the dropped table
+ tree2table.invalidate(treeName);
+ unstampableTrees.remove(treeName); // a table recreated later deserves a fresh stamp attempt
}
@Override
@@ -698,9 +1292,22 @@
final Connection con;
final ReadableTransactionImpl txr;
final WriteableTransactionTransactionImpl txw;
+ // The trees this import wrote: close() refreshes the statistics of these and only these,
+ // so rebuilding a single index does not gather statistics for the whole backend. A full
+ // import legitimately covers every tree - AbstractTwoPhaseImportStrategy.beforePhaseOne
+ // clears them all before the first record is written - including when the import is
+ // aborted, since close() runs from the try-with-resources of OnDiskMergeImporter.
+ final Set<TreeName> writtenTrees = ConcurrentHashMap.newKeySet();
+
+ // Set when the import failed or was cancelled. Its trees hold whatever the import got
+ // through before it stopped - beforePhaseOne cleared them all, so that can be nothing at
+ // all - and the operator is going to run it again, so there is nothing worth describing
+ // to the optimizer here: on oracle gathering those statistics is a full scan per table
+ // that would delay the report of a failure, or of a cancellation, by all of its duration.
+ volatile boolean aborted = false;
final Boolean isOpen;
-
+
public ImporterImpl() {
isOpen=getStorageStatus().isWorking();
if (!isOpen) {
@@ -720,26 +1327,46 @@
}
@Override
+ public void aborted() {
+ aborted = true;
+ }
+
+ @Override
public void close() {
try {
- con.commit();
- con.close();
+ try {
+ con.commit();
+ if (aborted) {
+ logger.debug(LocalizableMessage.raw("jdbc: import aborted: statistics of the trees it wrote are left alone"));
+ }else {
+ updateTableStatistics(con, writtenTrees);
+ }
+ } finally { // the pooled connection must be returned even when the commit or a statistics statement throws
+ try {
+ con.close();
+ } finally {
+ txw.stampSession.close();
+ }
+ }
} catch (SQLException e) {
throw new StorageRuntimeException(e);
- }
- if (!isOpen) {
- JDBCStorage.this.close();
+ } finally {
+ if (!isOpen) {
+ JDBCStorage.this.close();
+ }
}
}
-
+
@Override
public void clearTree(TreeName name) {
txw.clearTree(name);
+ writtenTrees.add(name);
}
-
+
@Override
public void put(TreeName treeName, ByteSequence key, ByteSequence value) {
txw.put(treeName, key, value);
+ writtenTrees.add(treeName);
}
@Override
--
Gitblit v1.10.0