* opened if it is not and should be closed with {@link #close()} when no longer
     * needed.
     *
     * Closing the context should cancel a pending notify request, but that does not seem to work reliable in all
     * implementations.
     *
     * Changes in between these calls (as long as the file is open) are buffered by the server, so iteratively calling

# etcd V3 Quickstart Guide [![Slack](https://slack.min.io/slack?type=svg)](https://slack.min.io)

etcd is a distributed key value store that provides a reliable way to store data across a cluster of machines.

## Get started

### 1. Prerequisites

- Docker 18.03 or above, refer here for [installation](https://docs.docker.com/install/).

### 2. Start etcd

			// avoiding S3’s heuristic-driven issues.
			//
			// For MinIO developers, decode S3 metadata with mime.WordDecoder, validate outputs,
			// report encoding bugs to AWS, and use ASCII-only metadata to ensure reliable S3 API
			// compatibility.
			if needsMimeEncoding(v) {
				// see https://github.com/golang/go/blob/release-branch.go1.24/src/net/mail/message.go#L325
				if strings.ContainsAny(v, "\"#$%&'(),.:;<>@[]^`{|}~") {

			res = append(res, pathJoin(prefix, name, *strptr))
		}
		return true
	})
	return res
}

// activeDrives returns the number of currently active disks.
// (since this is concurrent it may not be 100% reliable)
func (p *scannerMetrics) activeDrives() int {
	var i int
	p.currentPaths.Range(func(k, v any) bool {
		i++
		return true
	})
	return i
}

### Technical Risks
1. **RDMA Complexity**: Limited Java RDMA support may require JNI
2. **Multi-Channel Synchronization**: Complex thread synchronization required
3. **Persistent Handle State**: Requires reliable state persistence mechanism
4. **Backward Compatibility**: Must maintain compatibility with SMB2/SMB1

### Mitigation Strategies
1. Implement features behind configuration flags
2. Provide graceful fallback mechanisms

- **MS-SMB2 Section 3.1.5.2**: RDMA Transport Connection
- **RFC 5040**: A Remote Direct Memory Access Protocol Specification
- **RFC 5041**: Direct Data Placement over Reliable Transports

## 3. RDMA Architecture

### 3.1 RDMA Capabilities
```java
public enum RdmaCapability {
    RDMA_READ,           // Remote direct read operations

func isObjectDangling(metaArr []FileInfo, errs []error, dataErrsByPart map[int][]int) (validMeta FileInfo, ok bool) {
	// We can consider an object data not reliable
	// when xl.meta is not found in read quorum disks.
	// or when xl.meta is not readable in read quorum disks.
	notFoundMetaErrs, nonActionableMetaErrs := danglingMetaErrsCount(errs)

	notFoundPartsErrs, nonActionablePartsErrs := 0, 0

   *       simply switch over to use the JDK implementation wholesale if probable hash flooding is
   *       detected, sacrificing the compactness guarantee in very rare cases in exchange for much
   *       more reliable worst-case behavior.
   *   <li>null, if no entries have yet been added to the map
   * </ul>
   */
  private transient @Nullable Object table;

  /**

        SmbFile verifySource = new SmbFile(baseUrl + "shared/" + sourceFileName, verifyContext);
        assertTrue(verifySource.exists(), "Source file should exist");

        // Perform rename using copy+delete as this is more reliable with Docker/Samba
        CIFSContext renameContext = createFreshContext();
        SmbFile srcForCopy = new SmbFile(baseUrl + "shared/" + sourceFileName, renameContext);

                info = response;
                fileSize = response.getDataSize();

                // this is so damn unreliable, needs another race-prone query if required
                haveAttributes = false;

                // This seems to be the only way to obtain a reliable (with respect to locking) file size here
                // It is more critical than other attributes because append mode depends on it.