zstan commented on code in PR #13366:
URL: https://github.com/apache/ignite/pull/13366#discussion_r3728376399
##########
modules/calcite/src/main/java/org/apache/ignite/internal/processors/query/calcite/exec/ExecutionServiceImpl.java:
##########
@@ -581,6 +598,290 @@ private FieldsQueryCursor<List<?>>
executeDdl(RootQuery<Row> qry, DdlPlan plan)
}
}
+ /**
+ * Executes a {@code SELECT ... FOR UPDATE} plan.
+ *
+ * <ol>
+ * <li>Validates that the current transaction is PESSIMISTIC.</li>
+ * <li>Runs the inner SELECT with hidden key, value, and version columns
and materialises all rows.</li>
+ * <li>Builds cache entries from the hidden columns.</li>
+ * <li>Creates a savepoint, acquires pessimistic locks via {@code
lockTxEntries()},
+ * and releases the savepoint on success (or rolls back on
failure).</li>
+ * <li>Repeats the SELECT and lock attempt after a concurrent version
change while the deadline permits.</li>
+ * <li>Returns a cursor with only the user-visible columns (the appended
_KEY is stripped).</li>
+ * </ol>
+ */
+ private FieldsQueryCursor<List<?>> executeForUpdate(RootQuery<Row> qry,
SelectForUpdatePlan plan) {
+ GridNearTxLocal userTx = Commons.queryTransaction(qry.context(),
ctx.cache().context());
+
+ if (userTx == null || !userTx.pessimistic())
+ throw new IgniteSQLException(
+
IgniteResource.INSTANCE.selectForUpdateRequiresPessimisticTx().str(),
+ IgniteQueryErrorCode.UNSUPPORTED_OPERATION);
+
+ long waitMs = waitMillis(plan);
+
+ // Zero means that retries are limited only by the transaction or
query timeout.
+ long deadline = waitMs > 0
+ ? U.currentTimeMillis() + waitMs
+ : waitMs < 0 ? U.currentTimeMillis() : 0L;
+
+ RootQuery<Row> selectQry = qry;
+
+ while (true) {
+ FieldsQueryCursor<List<?>> cursor = tryExecuteForUpdate(selectQry,
plan, userTx, waitMs, deadline);
+
+ if (cursor != null)
+ return cursor;
+
+ if (deadline != 0 && U.currentTimeMillis() >= deadline) {
+ throw new IgniteSQLException(
+ IgniteResource.INSTANCE.selectForUpdateLockFailed().str(),
+ IgniteQueryErrorCode.CONCURRENT_UPDATE);
+ }
+
+ // The previous query has already been closed after
materialisation, so retry with a fresh root query.
+ selectQry = qry.retryQuery();
Review Comment:
misunderstand here - if we already obtain materialisation = results ? then -
cursor will return a bit upper. Seems here you need to write smth. like : "in
case of failed lock acquisition ... " and seems all we need here is just to
reset RootNode.state ? Use a constructor is ok but let\`s rename method like:
selectQry = qry.clone(); ?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
