* Email notifications sent after performing an action:
    * As connecting to an email server and sending an email tends to be "slow" (several seconds), you can return the response right away and send the email notification in the background.
* Processing data:
    * For example, let's say you receive a file that must go through a slow process, you can return a response of "Accepted" (HTTP 202) and process the file in the background.

TFCI_OUTPUT_DIR=build_output
TFCI_FIND_BIN=C:/tools/msys64/usr/bin/find.exe
TFCI_LIB_SUFFIX="-cpu-windows-x86_64"
# auditwheel is not supported for Windows
TFCI_WHL_AUDIT_ENABLE=0
TFCI_WHL_AUDIT_PLAT=0
# Tests are extremely slow at the moment
TFCI_WHL_BAZEL_TEST_ENABLE=0
TFCI_WHL_SIZE_LIMIT=450M
TFCI_WHL_SIZE_LIMIT_ENABLE=1
TFCI_WHL_IMPORT_TEST_ENABLE=1
TFCI_PYTHON_VERIFY_PIP_INSTALL_ARGS=""

	}

	st, err := gridConn.NewStream(ctx, grid.HandlerTrace, payload)
	if err != nil {
		return
	}
	st.Results(func(b []byte) error {
		select {
		case traceCh <- b:
		default:
			// Do not block on slow receivers.
			// Just recycle the buffer.
			grid.PutByteBuffer(b)
		}
		return nil
	})
}

func (client *peerRESTClient) doListen(ctx context.Context, listenCh chan<- []byte, v url.Values) {

	GlobalDeploymentID string
)

const (
	// DefaultIdleTimeout for idle inactive connections
	DefaultIdleTimeout = 30 * time.Second

	// DefaultReadHeaderTimeout for very slow inactive connections
	DefaultReadHeaderTimeout = 30 * time.Second

	// DefaultMaxHeaderBytes - default maximum HTTP header size in bytes.
	DefaultMaxHeaderBytes = 1 * humanize.MiByte
)

// and returns nil.
func osMkdirAll(dirPath string, perm os.FileMode, baseDir string) error {
	if baseDir != "" {
		if strings.HasPrefix(baseDir, dirPath) {
			return nil
		}
	}

	// Slow path: make sure parent exists and then call Mkdir for path.
	i := len(dirPath)
	for i > 0 && os.IsPathSeparator(dirPath[i-1]) { // Skip trailing path separator.
		i--
	}

	j := i

	}

	rulesMap := event.NewRulesMap(eventNames, pattern, event.TargetID{ID: mustGetUUID()})

	// Listen Publisher uses nonblocking publish and hence does not wait for slow subscribers.
	// Use buffered channel to take care of burst sends or slow w.Write()
	ch := make(chan event.Event, globalAPIConfig.getRequestsPoolCapacity())
	err := globalHTTPListen.Subscribe(mask, ch, ctx.Done(), func(ev event.Event) bool {

You could load it at the top level of the module/file, but that would also mean that it would **load the model** even if you are just running a simple automated test, then that test would be **slow** because it would have to wait for the model to load before being able to run an independent part of the code.

The exit code, the automatic closing of the `Session` in:

{* ../../docs_src/dependencies/tutorial013_an_py310.py ln[19:21] *}

...would be run after the the response finishes sending the slow data:

{* ../../docs_src/dependencies/tutorial013_an_py310.py ln[30:38] hl[31:33] *}

But as `generate_stream()` doesn't use the database session, it is not really necessary to keep the session open while sending the response.

    val extraParameters: String = "",
) {
    PER_COMMIT(
        displayName = "Performance Regression Test",
        timeout = 420,
        channel = "commits",
    ),
    PER_DAY(
        displayName = "Slow Performance Regression Test",
        timeout = 420,
        channel = "commits",
    ),
    PER_WEEK(
        displayName = "Performance Experiment",
        timeout = 420,
        channel = "experiments",

### Creating the `Settings` only once with `lru_cache` { #creating-the-settings-only-once-with-lru-cache }

Reading a file from disk is normally a costly (slow) operation, so you probably want to do it only once and then reuse the same settings object, instead of reading it for each request.

But every time we do:

```Python
Settings()
```