Commit 1115e29f authored Oct 15, 2020 by Josh Gao

adb: fix timeout-related test flakiness.

When running tests against cuttlefish, `adb root` sometimes flakes
because of a 6 second timeout in wait_for_device, which can get
triggered when adbd's restart gets throttled by init. The adb client
would notice that the socket went down, try to reconnect and fail
because adbd still hasn't started yet, and then wait up to 10 seconds to
reconnect, before which the `adb root` invocation would have failed due
to timeout. Increase the timeout to be long enough to handle this, and
also reduce the interval of reconnections to speed things up when we're
in this scenario.

Bug: http://b/170707786
Test: while true; do adb root; adb wait-for-device; adb unroot; adb wait-for-device; done
Change-Id: I140d721dfcfd635d45813a55990684a25e32228c

parent 3abb8b14

adb/client/commandline.cpp

+1 −1

Original line number	Diff line number	Diff line
		@@ -1149,7 +1149,7 @@ static bool adb_root(const char* command) {
		// If we were using a specific transport ID, there's nothing we can wait for.
		if (previous_id == 0) {
		adb_set_transport(previous_type, previous_serial, 0);
		wait_for_device("wait-for-device", 6000ms);
		wait_for_device("wait-for-device", 12000ms);
		}

		return true;

adb/transport.cpp

+2 −2

Original line number	Diff line number	Diff line
		@@ -129,8 +129,8 @@ class ReconnectHandler {
		};

		// Only retry for up to one minute.
		static constexpr const std::chrono::seconds kDefaultTimeout = 10s;
		static constexpr const size_t kMaxAttempts = 6;
		static constexpr const std::chrono::seconds kDefaultTimeout = 3s;
		static constexpr const size_t kMaxAttempts = 20;

		// Protects all members.
		std::mutex reconnect_mutex_;