From e0d528980f1987dd742aa104f4728031703007de Mon Sep 17 00:00:00 2001
From: Jibing-Li <64681310+Jibing-Li@users.noreply.github.com>
Date: Fri, 16 Dec 2022 09:36:35 +0800
Subject: [PATCH] [fix](multi catalog)Return emtpy block while external table
 scanner couldn't find the file (#14997)

FE file path cache for external table may out of date. In this case, BE may fail to find the not exist file from FE cache.
This pr is to handle this case: instead of throw an error message to the user, we return empty result set to the user.
---
 be/src/io/hdfs_file_reader.cpp             |  3 +++
 be/src/vec/exec/scan/scanner_scheduler.cpp | 12 +++++++++++-
 be/src/vec/exec/scan/vfile_scanner.cpp     |  3 +++
 3 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/be/src/io/hdfs_file_reader.cpp b/be/src/io/hdfs_file_reader.cpp
index 8c0e77e136..ba26154388 100644
--- a/be/src/io/hdfs_file_reader.cpp
+++ b/be/src/io/hdfs_file_reader.cpp
@@ -72,6 +72,9 @@ Status HdfsFileReader::open() {
 
     RETURN_IF_ERROR(HdfsFsCache::instance()->get_connection(_hdfs_params, &_fs_handle));
     _hdfs_fs = _fs_handle->hdfs_fs;
+    if (hdfsExists(_hdfs_fs, _path.c_str()) != 0) {
+        return Status::NotFound("{} not exists!", _path);
+    }
     _hdfs_file = hdfsOpenFile(_hdfs_fs, _path.c_str(), O_RDONLY, 0, 0, 0);
     if (_hdfs_file == nullptr) {
         if (_fs_handle->from_cache) {
diff --git a/be/src/vec/exec/scan/scanner_scheduler.cpp b/be/src/vec/exec/scan/scanner_scheduler.cpp
index f2170a9599..d45adc13b9 100644
--- a/be/src/vec/exec/scan/scanner_scheduler.cpp
+++ b/be/src/vec/exec/scan/scanner_scheduler.cpp
@@ -26,6 +26,7 @@
 #include "vec/core/block.h"
 #include "vec/exec/scan/vscanner.h"
 #include "vec/exprs/vexpr.h"
+#include "vfile_scanner.h"
 
 namespace doris::vectorized {
 
@@ -232,12 +233,21 @@ void ScannerScheduler::_scanner_scan(ScannerScheduler* scheduler, ScannerContext
         auto block = ctx->get_free_block(&get_free_block);
         status = scanner->get_block(state, block, &eos);
         VLOG_ROW << "VOlapScanNode input rows: " << block->rows() << ", eos: " << eos;
-        if (!status.ok()) {
+        // The VFileScanner for external table may try to open not exist files,
+        // Because FE file cache for external table may out of date.
+        if (!status.ok() && (typeid(*scanner) == typeid(doris::vectorized::VFileScanner) &&
+                             !status.is<ErrorCode::NOT_FOUND>())) {
             LOG(WARNING) << "Scan thread read VOlapScanner failed: " << status.to_string();
             // Add block ptr in blocks, prevent mem leak in read failed
             blocks.push_back(block);
             break;
         }
+        if (status.is<ErrorCode::NOT_FOUND>()) {
+            // The only case in this if branch is external table file delete and fe cache has not been updated yet.
+            // Set status to OK.
+            status = Status::OK();
+            eos = true;
+        }
 
         raw_bytes_read += block->bytes();
         num_rows_in_block += block->rows();
diff --git a/be/src/vec/exec/scan/vfile_scanner.cpp b/be/src/vec/exec/scan/vfile_scanner.cpp
index 1d67056920..1ad6928799 100644
--- a/be/src/vec/exec/scan/vfile_scanner.cpp
+++ b/be/src/vec/exec/scan/vfile_scanner.cpp
@@ -531,6 +531,9 @@ Status VFileScanner::_get_next_reader() {
         if (init_status.is<END_OF_FILE>()) {
             continue;
         } else if (!init_status.ok()) {
+            if (init_status.is<ErrorCode::NOT_FOUND>()) {
+                return init_status;
+            }
             return Status::InternalError("failed to init reader for file {}, err: {}", range.path,
                                          init_status.to_string());
         }