Method RecoverWriteError

be/src/runtime/tmp-file-mgr.cc:1824–1870

}

Status TmpFileGroup::RecoverWriteError(
    TmpWriteHandle* handle, const Status& write_status) {
  DCHECK(!write_status.ok());
  DCHECK(handle->file_ != nullptr);

  // We can't recover from cancellation or memory limit exceeded.
  if (write_status.IsCancelled() || write_status.IsMemLimitExceeded()) {
    return write_status;
  }

  // We don't recover the errors generated during spilling to a remote file.
  if (handle->file_->disk_type() != io::DiskFileType::LOCAL) {
    return write_status;
  }

  // Save and report the error before retrying so that the failure isn't silent.
  {
    lock_guard<SpinLock> lock(lock_);
    scratch_errors_.push_back(write_status);
    if (handle->file_->Blacklist(write_status.msg())) {
      DCHECK_LT(num_blacklisted_files_, tmp_files_.size());
      ++num_blacklisted_files_;
      if (num_blacklisted_files_ == tmp_files_.size()) {
        // Check if all errors are 'blacklistable'.
        bool are_all_blacklistable_errors = true;
        for (Status& err : scratch_errors_) {
          if (!ErrorConverter::IsBlacklistableError(err)) {
            are_all_blacklistable_errors = false;
            break;
          }
        }
        if (are_all_blacklistable_errors) spilling_disk_faulty_ = true;
      }
    }
  }

  // Do not retry cancelled writes or propagate the error, simply return CANCELLED.
  if (handle->is_cancelled_) return Status::CancelledInternal("TmpFileMgr write");

  TmpFile* tmp_file;
  int64_t file_offset;
  // Discard the scratch file range - we will not reuse ranges from a bad file.
  // Choose another file to try. Blacklisting ensures we don't retry the same file.
  // If this fails, the status will include all the errors in 'scratch_errors_'.
  RETURN_IF_ERROR(AllocateSpace(handle->on_disk_len(), &tmp_file, &file_offset));
  return handle->RetryWrite(io_ctx_.get(), tmp_file, file_offset);
}

Status TmpFileGroup::ScratchAllocationFailedStatus(
    const vector<int>& at_capacity_dirs) {