(cloudberry-pxf) 02/02: Fix: Add load/performance/extension test groups and align env/test expectations

[dj...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

djwang pushed a commit to branch merge-with-upstream
in repository https://gitbox.apache.org/repos/asf/cloudberry-pxf.git

commit 2b2959c32dec367ed5a0b7ded39d8ed0d8e1984f
Author: huluhuifeng <10...@qq.com>
AuthorDate: Wed Dec 31 07:25:55 2025 +0800

    Fix: Add load/performance/extension test groups and align env/test expectations
    
    - add bench_test and pxf_extension_test in run_tests.sh, plus matrix entries for bench and pxf_extension in CI
    - bump surefire heap to 4G to avoid OOM
    - update gpupgrade expected outputs to new PXF_HOME paths and JSON formatter error text
    - make ProtocolUtils/HiveBaseTest/JdbcHiveTest/OrcWriteTest/ParquetWriteTest more robust to env defaults (protocol, creds, hive JDBC URL)
    - keep MultiServerTest running under HDFS with a safe working directory fallback
    - set distribution key and INSERT pattern for performance test data load
---
 .../pxf/automation/components/hdfs/Hdfs.java       | 62 +++++++++++++++++++---
 1 file changed, 55 insertions(+), 7 deletions(-)

diff --git a/automation/src/main/java/org/greenplum/pxf/automation/components/hdfs/Hdfs.java b/automation/src/main/java/org/greenplum/pxf/automation/components/hdfs/Hdfs.java
index ac59a60b..1b422487 100755
--- a/automation/src/main/java/org/greenplum/pxf/automation/components/hdfs/Hdfs.java
+++ b/automation/src/main/java/org/greenplum/pxf/automation/components/hdfs/Hdfs.java
@@ -570,16 +570,64 @@ public class Hdfs extends BaseSystemObject implements IFSFunctionality {
         if (parent != null) {
             fs.mkdirs(parent);
         }
-        FSDataOutputStream out = fs.create(datapath, true,
-                bufferSize, replicationSize, blockSize);
 
-        DataOutputStream dos = out;
-        if (codec != null) {
-            dos = new DataOutputStream(codec.createOutputStream(out));
+        final int maxAttempts = 3;
+        try {
+            for (int attempt = 1; attempt <= maxAttempts; attempt++) {
+                FSDataOutputStream out = null;
+                DataOutputStream dos = null;
+                try {
+                    out = fs.create(datapath, true, bufferSize, replicationSize, blockSize);
+                    dos = out;
+                    if (codec != null) {
+                        dos = new DataOutputStream(codec.createOutputStream(out));
+                    }
+                    writeTableToStream(dos, dataTable, delimiter, encoding, newLine);
+                    return;
+                } catch (Exception e) {
+                    if (attempt >= maxAttempts || !isRetryableWriteException(e)) {
+                        throw e;
+                    }
+
+                    // Best-effort cleanup before retry (handles partially created files)
+                    try {
+                        if (dos != null) {
+                            dos.close();
+                        }
+                    } catch (Exception ignored) {
+                    }
+                    try {
+                        if (out != null) {
+                            out.close();
+                        }
+                    } catch (Exception ignored) {
+                    }
+                    try {
+                        fs.delete(datapath, false);
+                    } catch (Exception ignored) {
+                    }
+
+                    ReportUtils.report(report, getClass(),
+                            String.format("HDFS write failed (attempt %d/%d), retrying: %s", attempt, maxAttempts, e.getMessage()));
+                    Thread.sleep(2000L * attempt);
+                }
+            }
+        } finally {
+            ReportUtils.stopLevel(report);
         }
+    }
 
-        writeTableToStream(dos, dataTable, delimiter, encoding, newLine);
-        ReportUtils.stopLevel(report);
+    private boolean isRetryableWriteException(Exception e) {
+        if (e == null) {
+            return false;
+        }
+        String message = e.getMessage();
+        if (message == null) {
+            return false;
+        }
+        // Common transient failure on single-node HDFS when the only DataNode is briefly unavailable/blacklisted
+        return message.contains("could only be written to 0 of the 1 minReplication nodes")
+                || message.contains("node(s) are excluded in this operation");
     }
 
     public void appendTableToFile(String pathToFile, Table dataTable, String delimiter) throws Exception {


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cloudberry.apache.org
For additional commands, e-mail: commits-help@cloudberry.apache.org