GetDisks(poolIdx, setIdx int) ([]StorageAPI, error) // return the disks belonging to pool and set.
	SetDriveCounts() []int                              // list of erasure stripe size for each pool in order.

	// Healing operations.
	HealFormat(ctx context.Context, dryRun bool) (madmin.HealResultItem, error)

 * **Exchange** carries a single HTTP request/response pair.

 * **ExchangeFinder** chooses which connection carries each exchange. Where possible it will use the same connection for all exchanges in a single call. It prefers reusing pooled connections over establishing new connections.      

#### Per-Connection Locks

    @DisplayName("Should distinguish between pooled and non-pooled connections")
    void testPooledVsNonPooledConnections() {
        // When: Create pooled and non-pooled connections
        SmbTransportImpl pooled = pool.getSmbTransport(ctx, address, 445, false);
        SmbTransportImpl nonPooled = pool.getSmbTransport(ctx, address, 445, true);

        // Then: Pool should track pooled but not non-pooled

        CurlRequest request = new CurlRequest(Method.GET, "https://example.com");
        ForkJoinPool pool1 = new ForkJoinPool(2);
        ForkJoinPool pool2 = new ForkJoinPool(4);

        // Multiple threadPool calls should work
        request.threadPool(pool1);
        request.threadPool(pool2);

        assertNotNull(request);
    }

    @Test
    public void testOnConnectWithNull() {

Once the response has been received, the connection will be returned to the pool so it can be reused for a future request. Connections are evicted from the pool after a period of inactivity.

        urlQueueService.insert(urlQueue);

        // Poll should return the item
        final OpenSearchUrlQueue polled = urlQueueService.poll(sessionId);
        assertNotNull(polled);
        assertEquals("http://www.example.com/page1", polled.getUrl());
        assertEquals(sessionId, polled.getSessionId());

        // Second poll should return null (queue is empty)

 * Fix: Support the `PATCH` method.
 * Fix: Support request bodies on `DELETE` method.
 * Fix: Drop the `okhttp-protocols` module.
 * Internal: Replaced internal byte array buffers with pooled buffers ("OkBuffer").


## Version 1.3.0

_2014-01-11_

 * New: Support for "PATCH" HTTP method in client and MockWebServer.
 * Fix: Drop `Content-Length` header when redirected from POST to GET.

 * your HTTP calls. This is because each client holds its own connection pool and thread pools.
 * Reusing connections and threads reduces latency and saves memory. Conversely, creating a client
 * for each request wastes resources on idle pools.
 *
 * Use `new OkHttpClient()` to create a shared instance with the default settings:
 *
 * ```java
 * // The singleton HTTP client.

package bpool

import "sync"

// Pool is a single type sync.Pool with a few extra properties:
// If New is not set Get may return the zero value of T.
type Pool[T any] struct {
	New func() T
	p   sync.Pool
}

// Get will retuen a new T
func (p *Pool[T]) Get() T {
	v, ok := p.p.Get().(T)
	if ok {
		return v
	}
	if p.New == nil {
		var t T

docker volume prune -f || true
docker volume rm $(docker volume ls -q -f dangling=true) || true

# Stop two nodes, one of each pool, to check that all S3 calls work while quorum is still there
[ "${MODE}" == "pools" ] && docker-compose -f minio-${MODE}.yaml stop minio2
[ "${MODE}" == "pools" ] && docker-compose -f minio-${MODE}.yaml stop minio6

# Pause one node, to check that all S3 calls work while one node goes wrong