apache/hadoop · error · IOException

"Content-Length header is not provided by the server when tr

Error message

"Content-Length header is not provided by the server when trying to fetch " + url

What it means

IOException from EditLogFileInputStream.URLLog when the HTTP response for a remote edit log has no Content-Length header at all. The edit-log fetcher must know the stream length up front (length() feeds tail-position math and integrity checks), so a response without the header -- typically transfer-encoding: chunked -- is rejected instead of read.

Source

Thrown at hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/EditLogFileInputStream.java:509

              }
              
              if (connection.getResponseCode() != HttpURLConnection.HTTP_OK) {
                throw new HttpGetFailedException(
                    "Fetch of " + url +
                    " failed with status code " + connection.getResponseCode() +
                    "\nResponse message:\n" + connection.getResponseMessage(),
                    connection);
              }
        
              String contentLength = connection.getHeaderField(CONTENT_LENGTH);
              if (contentLength != null) {
                advertisedSize = Long.parseLong(contentLength);
                if (advertisedSize <= 0) {
                  throw new IOException("Invalid " + CONTENT_LENGTH + " header: " +
                      contentLength);
                }
              } else {
                throw new IOException(CONTENT_LENGTH + " header is not provided " +
                                      "by the server when trying to fetch " + url);
              }
        
              return connection.getInputStream();
            }
          });
    }

    @Override
    public long length() {
      return advertisedSize;
    }

    @Override
    public String getName() {
      return url.toString();
    }
  }

View on GitHub (pinned to 2add963021)

Solutions

Make the tailer talk directly to the NameNode: clear http.proxyHost for NN hosts (http.nonProxyHosts) or remove the LB from that path.
Confirm with curl -sv what the endpoint really returns; if the NN itself omits the header, check its version and upgrade.
Reconfigure the proxy/LB to buffer the response so the upstream Content-Length is preserved.

Example fix

# before: standby -> nginx -> active, response arrives chunked with no Content-Length
curl -sv http://nginx:9870/getimage?getedit=1&txid=42 -o /dev/null   # shows Transfer-Encoding: chunked
# after: point the tailer straight at the active NN
# (set the shared-edits/tailing URL to http://nn-active:9870 or add it to no_proxy)

Defensive patterns

Strategy: try-catch

Validate before calling

curl -sv --negotiate '<getedit URL>' -o /dev/null 2>&1 | egrep -i 'content-length|transfer-encoding'
# 'Transfer-Encoding: chunked' where the NN would send Content-Length = intermediary re-chunking

Try / catch

catch (IOException e) {
  if (e.getMessage() != null && e.getMessage().contains("header is not provided")) {
    // env/proxy issue, not data loss: remove the intermediary for this URL and retry
  } else { throw e; }
}

Prevention

Keep tailer-to-NameNode traffic direct; configure proxies/LBs to buffer so Content-Length survives.
After any network change, verify the journal endpoint's headers with curl.
Pin dfs.namenode.http-address to the real NN host, not a chunking VIP.

When it happens

Trigger: An intermediary (proxy/LB) re-chunking the GetImageServlet response; a very old or nonconforming server; SSL-terminating frontends that drop the header. The tailer opens the URL, getHeaderField('Content-Length') returns null, and the fetch fails immediately.

Common situations: Corporate proxy or nginx/HAProxy between standby and active NN; direct curl shows 'Transfer-Encoding: chunked' where the NN itself would send Content-Length; JVM proxy env vars accidentally applying to NN traffic.

Related errors

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/d6353045e25e27ad. Report an issue: GitHub.