Pavel Pereslegin created IGNITE-26927:
-----------------------------------------
Summary: Java thin. Sql. A race between close() and
fetchNextPage() in ResultSet may result in an error
Key: IGNITE-26927
URL: https://issues.apache.org/jira/browse/IGNITE-26927
Project: Ignite
Issue Type: Bug
Components: sql ai3, thin clients ai3
Reporter: Pavel Pereslegin
Consider the following example:
{code:java}
var rs = sql.executeAsync(null, "SELECT...").join();
rs.fetchNextPage(); // SQL_CURSOR_NEXT_PAGE
rs.close(); // SQL_CURSOR_CLOSE
{code}
In the above case, the SQL_CURSOR_CLOSE request is sent to the server before
the result of the SQL_CURSOR_NEXT_PAGE request is received.
The SQL_CURSOR_NEXT_PAGE request handler, if the last requested page removes
the client resource (see ClientSqlCursorNextPageRequest.process())
The SQL_CURSOR_CLOSE request handler may fail with the error "Failed to find
resource with id" if SQL_CURSOR_NEXT_PAGE has already removed this resource.
The problem is even more important for the synchronous API due to optimization,
which always requests the next data page in the background.
Reproducer:
{code:java}
@Test
void resourceNotFoundOnClose() {
IgniteSql sql = client().sql();
Statement stmt = sql.statementBuilder()
.query("SELECT * FROM TABLE(SYSTEM_RANGE(0, 1))") // 2 rows
.pageSize(1) // 1 row
.build();
for (int attempt = 0; attempt < 1000; attempt++) {
var rs = sql.execute(null, stmt); // preloads 2-nd page in the
background (see SyncResultSetAdapter.IteratorImpl constructor)
rs.close(); // Fails with "Failed to find resource"
}
}
{code}
Stacktrace:
{noformat}
Caused by: org.apache.ignite.internal.lang.IgniteInternalException:
IGN-CMN-65535 Failed to find resource with id: 23 TraceId:e40ee6a7
at
org.apache.ignite.client.handler.ClientResourceRegistry.remove(ClientResourceRegistry.java:95)
at
org.apache.ignite.client.handler.requests.sql.ClientSqlCursorCloseRequest.process(ClientSqlCursorCloseRequest.java:35)
at
org.apache.ignite.client.handler.ClientInboundMessageHandler.processOperation(ClientInboundMessageHandler.java:994)
at
org.apache.ignite.client.handler.ClientInboundMessageHandler.processOperationInternal(ClientInboundMessageHandler.java:1103)
{noformat}
We need to fix this issue properly.
Some thoughts worth considering:
* Completely ignoring the exception is bad, because we might miss another
resource leak problem.
* Removing background preloading in SyncResultSetAdapter is bad because it
negatively affects performance.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
