ahuston-0/hydra
1
0
Fork 0
Code Issues Pull Requests 1 Packages Projects Releases Wiki Activity

Temporarily disable machines after a connection failure

Browse Source
This commit is contained in:
Eelco Dolstra
2015-07-21 15:53:27 +02:00
parent 7e026d35f7
commit c18fb0ad74
4 changed files with 43 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
src/hydra-queue-runner/dispatcher.cc
View File

@ -36,9 +36,12 @@ void State::dispatcher()
bool keepGoing;
do {
system_time now = std::chrono::system_clock::now();
/* Copy the currentJobs field of each machine. This is
necessary to ensure that the sort comparator below is
an ordering. std::sort() can segfault if it isn't. */
an ordering. std::sort() can segfault if it isn't. Also
filter out temporarily disabled machines. */
struct MachineInfo
{
Machine::ptr machine;
@ -47,8 +50,15 @@ void State::dispatcher()
std::vector<MachineInfo> machinesSorted;
{
auto machines_(machines.lock());
for (auto & m : *machines_)
for (auto & m : *machines_) {
auto info(m.second->state->connectInfo.lock());
if (info->consecutiveFailures && info->disabledUntil > now) {
if (info->disabledUntil < sleepUntil)
sleepUntil = info->disabledUntil;
continue;
}
machinesSorted.push_back({m.second, m.second->state->currentJobs});
}
}
/* Sort the machines by a combination of speed factor and
@ -77,7 +87,6 @@ void State::dispatcher()
on it. Once we find such a pair, we restart the outer
loop because the machine sorting will have changed. */
keepGoing = false;
system_time now = std::chrono::system_clock::now();
for (auto & mi : machinesSorted) {
// FIXME: can we lose a wakeup if a builder exits concurrently?