Allow overriding UVM start timeout.

Currently the uvm.Start() method hardcodes 2 minute timeout value in the context that gets
used when making hvsocket connections to the UVM (for entropy, output handling & GCS)
. This timeout is good enough for general scenario but is very restrictive when debugging
uvm boot issues. This change allows overriding the default timeout via environment
variables.

Another option that was considered, was to use the deadline/timeout passed in the context
instead of hardcoding it. However, there is code in the Start method that explicitly
assumes that the parent context won't have a timeout(we could probably get around that
problem by creating a new context without a deadline). Another problem is that there are
lot of existing callers of the Start method who may or may not include a proper
timeout/deadline in the context.  If we suddenly start using that value, they might start
seeing unexpected timeouts
(for smaller context deadlines) or they may start noticing uvm start hanging (for longer
deadlines).

Signed-off-by: Amit Barve <[email protected]>

Author: ambarve

internal/timeout/timeout.go

@@ -47,6 +47,11 @@ var (
 
 	// TestDRetryLoop is the timeout for testd retry loop when onlining a SCSI disk in LCOW
 	TestDRetryLoop = defaultTimeoutTestdRetry
+
+	// This timeout is used for GCS connection after uvm boot as well as the entropy
+	// and log connection setup during uvm boot. This is different than the
+	// SystemStart timeout defined above.
+	GCSConnectionTimeout = 2 * time.Minute
 )
 
 func init() {
@@ -60,6 +65,7 @@ func init() {
 	ExternalCommandToStart = durationFromEnvironment("HCSSHIM_TIMEOUT_EXTERNALCOMMANDSTART", ExternalCommandToStart)
 	ExternalCommandToComplete = durationFromEnvironment("HCSSHIM_TIMEOUT_EXTERNALCOMMANDCOMPLETE", ExternalCommandToComplete)
 	TestDRetryLoop = durationFromEnvironment("HCSSHIM_TIMEOUT_TESTDRETRYLOOP", TestDRetryLoop)
+	GCSConnectionTimeout = durationFromEnvironment("HCSSHIM_TIMEOUT_GCSCONNECTION", GCSConnectionTimeout)
 }
 
 func durationFromEnvironment(env string, defaultValue time.Duration) time.Duration {

internal/uvm/start.go

@@ -26,6 +26,7 @@ import (
 	"github.com/Microsoft/hcsshim/internal/logfields"
 	"github.com/Microsoft/hcsshim/internal/protocol/guestrequest"
 	"github.com/Microsoft/hcsshim/internal/protocol/guestresource"
+	"github.com/Microsoft/hcsshim/internal/timeout"
 	"github.com/Microsoft/hcsshim/internal/uvm/scsi"
 )
 
@@ -157,7 +158,9 @@ func (uvm *UtilityVM) configureHvSocketForGCS(ctx context.Context) (err error) {
 func (uvm *UtilityVM) Start(ctx context.Context) (err error) {
 	// save parent context, without timeout to use in terminate
 	pCtx := ctx
-	ctx, cancel := context.WithTimeout(pCtx, 2*time.Minute)
+	ctx, cancel := context.WithTimeout(pCtx, timeout.GCSConnectionTimeout)
+	log.G(ctx).Debugf("using gcs connection timeout: %s\n", timeout.GCSConnectionTimeout)
+
 	g, gctx := errgroup.WithContext(ctx)
 	defer func() {
 		_ = g.Wait()