What causes "Not all bytes of message (READY=1) sent to systemd socket (sent {})" in elastic/elasticsearch?

Check that no other code closes file descriptors that might alias the systemd socket fd.

elastic/elasticsearch · warning · RuntimeException

Not all bytes of message (READY=1) sent to systemd socket (s

Error message

Not all bytes of message (READY=1) sent to systemd socket (sent {})

What it means

Thrown as a RuntimeException when libc.send() returns a byte count less than the message length. For Unix datagram sockets this is unusual (datagrams are atomic), but it can occur if the socket was closed mid-send or if the kernel truncated the datagram. The error includes the number of bytes actually sent.

Source

Thrown at libs/native/src/main/java/org/elasticsearch/nativeaccess/Systemd.java:114

                    if (error != null) {
                        error.addSuppressed(e);
                        throw error;
                    } else {
                        throw e;
                    }
                }
            } else if (error != null) {
                throw error;
            }
        }
    }

    private void throwOrLog(String message, boolean warnOnError) {
        if (warnOnError) {
            logger.warn(message);
        } else {
            logger.error(message);
            throw new RuntimeException(message);
        }
    }
}

View on GitHub (pinned to db6a809a66)

Solutions

Check that no other code closes file descriptors that might alias the systemd socket fd.
Keep notification messages short (READY=1 is only 7 bytes; this error is very rare for it).
Wrap notify_ready() in try-catch as partial sends are unrecoverable.
If using a custom notification state string, ensure it does not exceed the socket buffer.

Example fix

// before
systemd.notify_ready();

// after
try {
    systemd.notify_ready();
} catch (RuntimeException e) {
    logger.warn("Partial sd_notify send; systemd may not have received READY=1", e);
}

Defensive patterns

Strategy: try-catch

Validate before calling

// Partial sends on SOCK_DGRAM are not predictable; use try-catch.
// Verify the service unit Type=notify and keep messages short.

Try / catch

try {
    systemd.notify_ready();
} catch (RuntimeException e) {
    logger.warn("systemd READY=1 partial send; continuing startup", e);
}

Prevention

Do not share or close file descriptors that may alias the internal systemd socket fd.
Keep notification state strings minimal.
Treat sd_notify as best-effort; never let it block startup.

When it happens

Trigger: Calling systemd.notify_ready() where send() returns a partial write. This is rare for SOCK_DGRAM but can happen if the socket fd is invalidated between connect and send, or if the message exceeds the maximum datagram size.

Common situations: Concurrent close of the socket fd by another thread. Extremely long notification strings exceeding SO_SNDBUF. Kernel or driver bug truncating datagrams. Unlikely but possible under heavy signal load interrupting the syscall.

Related errors

AI-assisted analysis of elastic/elasticsearch@db6a809a66 (2026-08-12). Data as JSON: /api/errors/0cc1b605610db7ba. Report an issue: GitHub.