Author: challngr
Date: Mon Aug 24 19:25:05 2015
New Revision: 1697485
URL: http://svn.apache.org/r1697485
Log:
UIMA-4569 RM should report allocations in GB instead of shares.
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/JobManagerConverter.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/IRmJob.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ISchedulerMain.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Machine.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodePool.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodepoolScheduler.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ResourceClass.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/RmJob.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Scheduler.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Share.java
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/User.java
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/RmStateDuccEvent.java
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/IRmJobState.java
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/Resource.java
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/RmJobState.java
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/JobFragment.java
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/Markup.java
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/NodeViz.java
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/VisualizedHost.java
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/JobManagerConverter.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/JobManagerConverter.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/JobManagerConverter.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/JobManagerConverter.java
Mon Aug 24 19:25:05 2015
@@ -124,7 +124,6 @@ public class JobManagerConverter
{
String methodName = "blacklistJob";
- int order = scheduler.calcShareOrder(memory);
Map<DuccId, IResource> all_shares = null;
Map<DuccId, IResource> shrunken_shares = null;
Map<DuccId, IResource> expanded_shares = null;
@@ -137,16 +136,18 @@ public class JobManagerConverter
// first time - everything must go
IDuccProcessMap pm = ((IDuccWorkExecutable)job).getProcessMap();
+ int quantum = 0;
for ( IDuccProcess proc : pm.values() ) { // build up Shares
from the incoming state
NodeIdentity ni = proc.getNodeIdentity();
Machine m = scheduler.getMachine(ni);
int share_order = 1;
+ quantum = m.getQuantum();
if ( m != null ) {
if ( proc.isActive() || (proc.getProcessState() ==
ProcessState.Undefined) ) {
logger.info(methodName, job.getDuccId(), "blacklist",
proc.getDuccId(),
"state", proc.getProcessState(), "isActive",
proc.isActive(), "isComplete", proc.isComplete());
- m.blacklist(job.getDuccId(), proc.getDuccId(), order);
+ m.blacklist(job.getDuccId(), proc.getDuccId(), memory);
if ( evict ) {
share_order = m.getShareOrder(); // best guess
Resource r = new Resource(proc.getDuccId(),
proc.getNode(), false, share_order, 0);
@@ -162,7 +163,7 @@ public class JobManagerConverter
}
if ( evict && (shrunken_shares.size() > 0) ) {
- RmJobState rjs = new RmJobState(job.getDuccId(), all_shares,
shrunken_shares, expanded_shares);
+ RmJobState rjs = new RmJobState(job.getDuccId(), quantum,
all_shares, shrunken_shares, expanded_shares);
rjs.setDuccType(job.getDuccType());
blacklistedResources.put(job.getDuccId(), rjs); // to
tell OR
}
@@ -1157,7 +1158,7 @@ public class JobManagerConverter
}
}
- RmJobState rjs = new RmJobState(j.getId(), all_shares,
shrunken_shares, expanded_shares);
+ RmJobState rjs = new RmJobState(j.getId(),
(j.getShareQuantum() >> 20) * (j.getShareOrder()), all_shares, shrunken_shares,
expanded_shares);
rjs.setDuccType(j.getDuccType());
rjs.setReason(j.getReason());
rmJobState.put(j.getId(), rjs);
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/IRmJob.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/IRmJob.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/IRmJob.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/IRmJob.java
Mon Aug 24 19:25:05 2015
@@ -49,6 +49,11 @@ public interface IRmJob
public void setJobName(String name);
/**
+ * Asks it's resource class what the share quantum is that it is being
scheduled under.
+ */
+ public int getShareQuantum();
+
+ /**
* Save ref to the class we are in, and init class-based structures.
*/
public void setResourceClass(ResourceClass cl);
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ISchedulerMain.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ISchedulerMain.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ISchedulerMain.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ISchedulerMain.java
Mon Aug 24 19:25:05 2015
@@ -62,7 +62,7 @@ public interface ISchedulerMain
IRmJob getJob(DuccId id);
Share getShare(DuccId id);
Machine getMachine(NodeIdentity id);
- int calcShareOrder(long mem); // UIMA-4142 expose to
JobManagerConverter
+ int calcShareOrder(IRmJob j); // UIMA-4142 expose to
JobManagerConverter
void queryMachines();
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Machine.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Machine.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Machine.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Machine.java
Mon Aug 24 19:25:05 2015
@@ -39,7 +39,7 @@ public class Machine
private int hbcounter = 0; // heartbeat counter
private long memory; // in Kb
- private int share_order = 0;
+ private int share_order = 1;
private NodePool nodepool;
@@ -94,13 +94,26 @@ public class Machine
return node;
}
+ /**
+ * Return the quantum for this machine - the actual amount allocated in a
single share on this machine.
+ */
+ public int getQuantum()
+ {
+ return (int) memory / share_order;
+ }
+
// UIMA-4142
// Black list some number of shres for a specific job and proc. This
reduces the number of
// schedulable shares until they are whitelisted.
- public synchronized void blacklist(DuccId jobid, DuccId procid, int
nshares)
+ //public synchronized void blacklist(DuccId jobid, DuccId procid, int
nshares)
+ public synchronized void blacklist(DuccId jobid, DuccId procid, long
jobmem)
{
String methodName = "blacklist";
-
+
+ int q = getQuantum();
+ int nshares = (int) jobmem / q;
+ if ( jobmem % q > 0 ) nshares++;
+
if ( ! blacklistedWork.containsKey(procid) ) { //
already condemned?
if ( nshares == -1 ) nshares = share_order; //
whole machine - reservations
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodePool.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodePool.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodePool.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodePool.java
Mon Aug 24 19:25:05 2015
@@ -1,3 +1,4 @@
+
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
@@ -19,6 +20,7 @@
package org.apache.uima.ducc.rm.scheduler;
import java.util.ArrayList;
+import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashMap;
@@ -42,6 +44,7 @@ class NodePool
int depth;
int updated = 0;
int search_order = 100;
+ int share_quantum = 0;
EvictionPolicy evictionPolicy = EvictionPolicy.SHRINK_BY_MACHINE;
@@ -92,7 +95,7 @@ class NodePool
// Indexed by available free shares, the specific machines with the
indicated free space
HashMap<Integer, Map<Node, Machine>> virtualMachinesByOrder = new
HashMap<Integer, Map<Node, Machine>>(); // UIMA-4142
- static int maxorder = 0;
+ GlobalOrder maxorder = null;
// NodePool(NodePool parent, String id, EvictionPolicy ep, int order)
// {
@@ -103,7 +106,7 @@ class NodePool
// this.order = order;
// }
- NodePool(NodePool parent, String id, Map<String, String> nodes,
EvictionPolicy ep, int depth, int search_order)
+ NodePool(NodePool parent, String id, Map<String, String> nodes,
EvictionPolicy ep, int depth, int search_order, int share_quantum)
{
String methodName = "NodePool.<init>";
this.parent = parent;
@@ -116,6 +119,13 @@ class NodePool
this.evictionPolicy = ep;
this.depth = depth;
this.search_order = search_order;
+ this.share_quantum = share_quantum;
+
+ if ( parent == null ) {
+ maxorder = new GlobalOrder();
+ } else {
+ maxorder = parent.getGlobalOrder();
+ }
}
void addResourceClass(ResourceClass cl)
@@ -133,6 +143,11 @@ class NodePool
return id;
}
+ int getShareQuantum()
+ {
+ return share_quantum;
+ }
+
int getDepth()
{
return depth;
@@ -481,22 +496,33 @@ class NodePool
return cln;
}
- public static int getMaxOrder()
+ public GlobalOrder getGlobalOrder()
{
- return maxorder; // must always be the same for parent
and all children
+ return maxorder;
}
- public static int[] makeArray() // common static code because
getting this right everywhere is painful
+ public void updateMaxOrder(int order)
{
- return new int[getArraySize()];
+ maxorder.update(order);
+ }
+
+ public int getMaxOrder()
+ {
+ return maxorder.getOrder(); // must always be the same
for parent and all children
}
- public static int getArraySize()
+ public int getArraySize()
{
- return getMaxOrder() + 1; // a bit bigger, because we're
1-indexed for easier counting
- // same for parent and children
+ return getMaxOrder() + 1; // a bit bigger,
because we're 1-indexed for easier counting
+ // same for parent and
children
}
+ public int[] makeArray() // common static code
because getting this right everywhere is painful
+ {
+ return new int[getArraySize()];
+ }
+
+
int getSearchOrder()
{
return this.search_order;
@@ -637,7 +663,7 @@ class NodePool
// init nSharesByorder to the sum of 'n and 'v MachinesByOrder
System.arraycopy(nMachinesByOrder, 0, nSharesByOrder, 0, len);
- for ( int i = 0; i < maxorder + 1; i++ ) {
+ for ( int i = 0; i < getMaxOrder() + 1; i++ ) {
nSharesByOrder[i] += vMachinesByOrder[i];
}
@@ -715,6 +741,11 @@ class NodePool
String methodName = "rearrangeVirtual";
if ( allMachines.containsKey(m.key()) ) {
int v_order = m.getVirtualShareOrder();
+
+ if ( v_order < 0 ) {
+ int stop_here = 1;
+ stop_here++;
+ }
int r_order = m.getShareOrder();
logger.trace(methodName, null, m.getId(), "order", order,
"v_order", v_order, "r_order", r_order);
@@ -783,14 +814,16 @@ class NodePool
// reset code so it matches the cycles better. otoh, this isn't
a performance-intensive
// scheduler so do we care?
//
- maxorder = Math.max(order, maxorder);
- nSharesByOrder = new int[maxorder + 1];
- nMachinesByOrder = new int[maxorder + 1];
- vMachinesByOrder = new int[maxorder + 1];
+ updateMaxOrder(order);
+ logger.info(methodName, null, "Nodepool:", id, "Maxorder set to",
getMaxOrder());
+
+ nSharesByOrder = makeArray();
+ nMachinesByOrder = makeArray();
+ vMachinesByOrder = makeArray();
//nFreeSharesByOrder = new int[maxorder + 1];
//neededByOrder = new int[maxorder + 1];
- nPendingByOrder = new int[maxorder + 1];
+ nPendingByOrder = makeArray();
// UIMA-4142 Must set vMachinesByOrder and virtualMachinesByOrder
independently of
// machinesByOrder because blacklisting can cause v_order !=
r_order
@@ -922,7 +955,7 @@ class NodePool
*/
NodePool createSubpool(String className, Map<String, String> names, int
order)
{
- NodePool np = new NodePool(this, className, names, evictionPolicy,
depth + 1, order);
+ NodePool np = new NodePool(this, className, names, evictionPolicy,
depth + 1, order, share_quantum);
children.put(className, np);
return np;
}
@@ -965,8 +998,7 @@ class NodePool
{
String methodName = "nodeArrives";
- maxorder = Math.max(order, maxorder);
-
+ updateMaxOrder(order);
for ( NodePool np : children.values() ) {
if ( np.containsPoolNode(node) ) {
@@ -1211,7 +1243,7 @@ class NodePool
{
int order = j.getShareOrder();
int ret = 0;
- for ( int i = order; i < maxorder; i++ ) {
+ for ( int i = order; i < getMaxOrder(); i++ ) {
if ( machinesByOrder.containsKey(order) ) {
ret += machinesByOrder.get(order).size();
}
@@ -1304,7 +1336,7 @@ class NodePool
int rem = 0;
int low = order;
- while ( (given < nrequested ) && ( low <= maxorder ) ) {
+ while ( (given < nrequested ) && ( low <= getMaxOrder() ) ) {
int avail = vMachinesByOrder[low] + nMachinesByOrder[low];
if ( avail > 0 ) {
@@ -1507,126 +1539,126 @@ class NodePool
return null; // found
nothing, heck
}
- // private void doEvictions(int[] neededByOrder, HashMap<Integer,
HashMap<IRmJob, IRmJob>> candidates, boolean force)
- // {
+ private void doEvictions(int[] neededByOrder, HashMap<Integer,
HashMap<IRmJob, IRmJob>> candidates, boolean force)
+ {
+
+ for ( int nbo = getMaxOrder(); nbo > 0; nbo-- ) {
+
+ if ( neededByOrder[nbo] == 0 ) {
// these are N-shares
+ continue;
+ }
+ for ( int oo = getMaxOrder(); oo > 0; oo-- ) {
+ HashMap<IRmJob, IRmJob> jobs = candidates.get(oo);
+ if ( jobs == null ) {
+ continue;
+ }
+
+ Iterator<IRmJob> iter = jobs.values().iterator();
// he has something to give. is it enough?
+ while ( iter.hasNext() && (neededByOrder[nbo] > 0) ) {
+ IRmJob j = iter.next();
+ int loss = 0;
+
+ switch ( evictionPolicy ) {
+ case SHRINK_BY_MACHINE:
+ // minimize fragmentation
+ loss =
j.shrinkByOrderByMachine(neededByOrder[nbo], nbo, force, this); // pass in
number of N-shares of given order that we want
+
// returns number of quantum shares it had to
relinquish
+ break;
+ case SHRINK_BY_INVESTMENT:
+ // minimize lost work
+ loss = j.shrinkByInvestment(neededByOrder[nbo],
nbo, force, this); // pass in number of N-shares of given order that we want
+
// returns number of quantum shares it had to
relinquish
+ break;
+ }
- // for ( int nbo = maxorder; nbo > 0; nbo-- ) {
+ neededByOrder[nbo] -= loss;
+ neededByOrder[0] -= loss;
+ nPendingByOrder[oo] += loss;
- // if ( neededByOrder[nbo] == 0 ) {
// these are N-shares
- // continue;
- // }
- // for ( int oo = maxorder; oo > 0; oo-- ) {
- // HashMap<IRmJob, IRmJob> jobs = candidates.get(oo);
- // if ( jobs == null ) {
- // continue;
- // }
-
- // Iterator<IRmJob> iter = jobs.values().iterator();
// he has something to give. is it enough?
- // while ( iter.hasNext() && (neededByOrder[nbo] > 0) ) {
- // IRmJob j = iter.next();
- // int loss = 0;
-
- // switch ( evictionPolicy ) {
- // case SHRINK_BY_MACHINE:
- // // minimize fragmentation
- // loss =
j.shrinkByOrderByMachine(neededByOrder[nbo], nbo, force, this); // pass in
number of N-shares of given order that we want
- //
// returns number of quantum shares it had to
relinquish
- // break;
- // case SHRINK_BY_INVESTMENT:
- // // minimize lost work
- // loss = j.shrinkByInvestment(neededByOrder[nbo],
nbo, force, this); // pass in number of N-shares of given order that we want
- //
// returns number of quantum shares it had to
relinquish
- // break;
- // }
-
- // neededByOrder[nbo] -= loss;
- // neededByOrder[0] -= loss;
- // nPendingByOrder[oo] += loss;
-
- // if ( j.countNShares() == 0 ) {
// nothing left? don't look here any more
- // iter.remove();
- // }
- // }
-
- // }
- // }
- // }
-
- // /**
- // * Here we tell the NP how much we need cleared up. It will look
around and try to do that.
- // * @deprecated No longer used, the doEvictions code in
NodepoolScheduler handles evictions by itself.
- // * Keeping this for a while for reference. UIMA-4275
- // */
- // void doEvictionsByMachine(int [] neededByOrder, boolean force)
- // {
- // String methodName = "doEvictions";
- // //
- // // Collect losers that are also squatters, by order, and try them
first
- // //
- // String type;
- // type = force ? "forced" : "natural";
-
- // logger.debug(methodName, null, getId(), "NeededByOrder", type, "on
entrance eviction", Arrays.toString(neededByOrder));
-
- // for ( NodePool np : getChildrenDescending() ) {
- // logger.info(methodName, null, "Recurse to", np.getId(), "from",
getId(), "force:", force);
- // np.doEvictionsByMachine(neededByOrder, force);
- // logger.info(methodName, null, "Recurse from", np.getId(),
"proceed with logic for", getId(), "force", force);
- // }
-
- // //
- // // Adjust neededByOrder to reflect the number of shares that need
to be preempted by subtracting the
- // // number of shares that already are free
- // //
- // for ( int nbo = maxorder; nbo > 0; nbo-- ) {
- // // UIMA-4065 - I think that subtracting
countPendingSharesByOrder() amounts to double counting because it
- // // will reflect any evictions from the depth-first
recursion. Instead, we would subtract only
- // // our own shares.
- // //
- // // int needed = Math.max(0, neededByOrder[nbo] -
countNSharesByOrder(nbo) - countPendingSharesByOrder(nbo));
- // int needed = Math.max(0, neededByOrder[nbo] -
countNSharesByOrder(nbo) - nPendingByOrder[nbo]);
- // neededByOrder[nbo] = needed;
- // neededByOrder[0] += needed;
- // }
-
- // logger.debug(methodName, null, getId(), "NeededByOrder", type,
"after adjustments for pending eviction:", Arrays.toString(neededByOrder));
-
- // HashMap<Integer, HashMap<IRmJob, IRmJob>> squatters = new
HashMap<Integer, HashMap<IRmJob, IRmJob>>();
- // HashMap<Integer, HashMap<IRmJob, IRmJob>> residents = new
HashMap<Integer, HashMap<IRmJob, IRmJob>>();
-
- // for ( Share s : allShares.values() ) {
- // HashMap<Integer, HashMap<IRmJob, IRmJob>> map = null;
- // boolean is_candidate = force ? s.isForceable() :
s.isPreemptable();
- // if ( is_candidate ) {
- // IRmJob j = s.getJob();
- // ResourceClass rc = j.getResourceClass();
- // if ( rc.getNodepoolName().equals(id) ) {
- // map = residents;
- // } else {
- // map = squatters;
- // }
-
- // int order = j.getShareOrder();
- // HashMap<IRmJob, IRmJob> jmap = null;
- // if ( map.containsKey(order) ) {
- // jmap = map.get(order);
- // } else {
- // jmap = new HashMap<IRmJob, IRmJob>();
- // map.put(order, jmap);
- // }
- // jmap.put(j, j);
- // }
- // }
-
- // doEvictions(neededByOrder, squatters, force);
- // logger.debug(methodName, null, getId(), "NeededByOrder", type,
"after eviction of squatters:", Arrays.toString(neededByOrder));
- // if ( neededByOrder[0] <= 0 ) {
- // return;
- // }
-
- // doEvictions(neededByOrder, residents, force);
- // logger.debug(methodName, null, getId(), "NeededByOrder", type,
"after eviction of residents:", Arrays.toString(neededByOrder));
- // }
+ if ( j.countNShares() == 0 ) {
// nothing left? don't look here any more
+ iter.remove();
+ }
+ }
+
+ }
+ }
+ }
+
+ /**
+ * Here we tell the NP how much we need cleared up. It will look around
and try to do that.
+ * @deprecated No longer used, the doEvictions code in NodepoolScheduler
handles evictions by itself.
+ * Keeping this for a while for reference. UIMA-4275
+ */
+ void doEvictionsByMachine(int [] neededByOrder, boolean force)
+ {
+ String methodName = "doEvictions";
+ //
+ // Collect losers that are also squatters, by order, and try them first
+ //
+ String type;
+ type = force ? "forced" : "natural";
+
+ logger.debug(methodName, null, getId(), "NeededByOrder", type, "on
entrance eviction", Arrays.toString(neededByOrder));
+
+ for ( NodePool np : getChildrenDescending() ) {
+ logger.info(methodName, null, "Recurse to", np.getId(), "from",
getId(), "force:", force);
+ np.doEvictionsByMachine(neededByOrder, force);
+ logger.info(methodName, null, "Recurse from", np.getId(), "proceed
with logic for", getId(), "force", force);
+ }
+
+ //
+ // Adjust neededByOrder to reflect the number of shares that need to
be preempted by subtracting the
+ // number of shares that already are free
+ //
+ for ( int nbo = getMaxOrder(); nbo > 0; nbo-- ) {
+ // UIMA-4065 - I think that subtracting
countPendingSharesByOrder() amounts to double counting because it
+ // will reflect any evictions from the depth-first
recursion. Instead, we would subtract only
+ // our own shares.
+ //
+ // int needed = Math.max(0, neededByOrder[nbo] -
countNSharesByOrder(nbo) - countPendingSharesByOrder(nbo));
+ int needed = Math.max(0, neededByOrder[nbo] -
countNSharesByOrder(nbo) - nPendingByOrder[nbo]);
+ neededByOrder[nbo] = needed;
+ neededByOrder[0] += needed;
+ }
+
+ logger.debug(methodName, null, getId(), "NeededByOrder", type, "after
adjustments for pending eviction:", Arrays.toString(neededByOrder));
+
+ HashMap<Integer, HashMap<IRmJob, IRmJob>> squatters = new
HashMap<Integer, HashMap<IRmJob, IRmJob>>();
+ HashMap<Integer, HashMap<IRmJob, IRmJob>> residents = new
HashMap<Integer, HashMap<IRmJob, IRmJob>>();
+
+ for ( Share s : allShares.values() ) {
+ HashMap<Integer, HashMap<IRmJob, IRmJob>> map = null;
+ boolean is_candidate = force ? s.isForceable() : s.isPreemptable();
+ if ( is_candidate ) {
+ IRmJob j = s.getJob();
+ ResourceClass rc = j.getResourceClass();
+ if ( rc.getNodepoolName().equals(id) ) {
+ map = residents;
+ } else {
+ map = squatters;
+ }
+
+ int order = j.getShareOrder();
+ HashMap<IRmJob, IRmJob> jmap = null;
+ if ( map.containsKey(order) ) {
+ jmap = map.get(order);
+ } else {
+ jmap = new HashMap<IRmJob, IRmJob>();
+ map.put(order, jmap);
+ }
+ jmap.put(j, j);
+ }
+ }
+
+ doEvictions(neededByOrder, squatters, force);
+ logger.debug(methodName, null, getId(), "NeededByOrder", type, "after
eviction of squatters:", Arrays.toString(neededByOrder));
+ if ( neededByOrder[0] <= 0 ) {
+ return;
+ }
+
+ doEvictions(neededByOrder, residents, force);
+ logger.debug(methodName, null, getId(), "NeededByOrder", type, "after
eviction of residents:", Arrays.toString(neededByOrder));
+ }
int findShares( IRmJob j )
@@ -1938,4 +1970,28 @@ class NodePool
}
}
+ class GlobalOrder
+ {
+ int maxorder = 0;
+
+ GlobalOrder()
+ {
+ this.maxorder = 0;
+ }
+
+ synchronized void reset()
+ {
+ this.maxorder = 0;
+ }
+
+ synchronized void update(int order)
+ {
+ this.maxorder = Math.max(maxorder, order);
+ }
+
+ synchronized int getOrder()
+ {
+ return maxorder;
+ }
+ }
}
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodepoolScheduler.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodepoolScheduler.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodepoolScheduler.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/NodepoolScheduler.java
Mon Aug 24 19:25:05 2015
@@ -106,15 +106,18 @@ public class NodepoolScheduler
}
fragmentationThreshold =
SystemPropertyResolver.getIntProperty("ducc.rm.fragmentation.threshold",
fragmentationThreshold);
- scheduling_quantum =
SystemPropertyResolver.getIntProperty("ducc.rm.share.quantum",
scheduling_quantum);
do_defragmentation =
SystemPropertyResolver.getBooleanProperty("ducc.rm.defragmentation",
do_defragmentation);
use_global_allotment =
SystemPropertyResolver.getBooleanProperty("ducc.rm.use_global_allotment",
use_global_allotment);
global_allotment =
SystemPropertyResolver.getIntProperty("ducc.rm.global_allotment",
global_allotment);
}
+ /**
+ * Sets the top-level np for this scheduler
+ */
public void setNodePool(NodePool np)
{
this.globalNodepool = np;
+ this.scheduling_quantum = np.getShareQuantum() >> 20; // to GB from
KB
}
public void setEvictionPolicy(EvictionPolicy ep)
@@ -385,8 +388,8 @@ public class NodepoolScheduler
{
String methodName = "apportion_qshares";
boolean shares_given = false;
- int maxorder = NodePool.getMaxOrder();
- int[] nshares = NodePool.makeArray(); // nshares
+ int maxorder = globalNodepool.getMaxOrder();
+ int[] nshares = globalNodepool.makeArray(); // nshares
if ( entities.size() == 0 ) return;
@@ -408,12 +411,12 @@ public class NodepoolScheduler
StringBuffer enames = null;
StringBuffer eweights = null;
logger.info(methodName, null, descr, "RmCounter Start");
- logger.info(methodName, null, descr, "maxorder = ",
NodePool.getMaxOrder());
+ logger.info(methodName, null, descr, "maxorder = ",
globalNodepool.getMaxOrder());
enames = new StringBuffer();
eweights = new StringBuffer();
for ( IEntity e : working ) {
- int[] gbo = NodePool.makeArray();
+ int[] gbo = globalNodepool.makeArray();
e.setGivenByOrder(gbo);
given_by_order.put(e, gbo);
@@ -1334,7 +1337,7 @@ public class NodepoolScheduler
// The job passes; give the job a count
//
logger.info(methodName, j.getId(), "+++++ nodepool", np.getId(),
"class", rc.getName(), "order", order, "shares", nSharesToString(granted,
order));
- int[] gbo = NodePool.makeArray();
+ int[] gbo = globalNodepool.makeArray();
gbo[order] = granted; // what we get
j.setGivenByOrder(gbo);
@@ -1359,7 +1362,7 @@ public class NodepoolScheduler
// already accounted for as well, since it is a non-preemptable
share
logger.info(methodName, j.getId(), "[stable]", "assigned",
j.countNShares(), "processes, ",
(j.countNShares() * j.getShareOrder()), "QS");
- int[] gbo = NodePool.makeArray();
+ int[] gbo = globalNodepool.makeArray();
gbo[j.getShareOrder()] = 1; // must set the
allocation so eviction works right
j.setGivenByOrder(gbo);
@@ -1393,7 +1396,7 @@ public class NodepoolScheduler
// The job passes. Assign it a count and get on with life ...
//
logger.info(methodName, j.getId(), "+++++ nodepool", np.getId(),
"class", rc.getName(), "order", order, "shares", nSharesToString(1, order));
- int[] gbo = NodePool.makeArray();
+ int[] gbo = globalNodepool.makeArray();
gbo[order] = 1;
j.setGivenByOrder(gbo);
@@ -1542,7 +1545,7 @@ public class NodepoolScheduler
// The job passes; give the job a count
//
logger.info(methodName, j.getId(), "Request is granted a machine for
reservation.");
- int[] gbo = NodePool.makeArray();
+ int[] gbo =globalNodepool.makeArray();
int order = j.getShareOrder(); // memory, coverted to order, so we
can find stuff
gbo[order] = freeable + j.countNShares(); // account for new stuff
plus what it already has
j.setGivenByOrder(gbo);
@@ -1560,7 +1563,7 @@ public class NodepoolScheduler
logger.info(methodName, j.getId(), "[stable]", "assigned",
j.countNShares(), "processes, ",
(j.countNShares() * j.getShareOrder()), "QS");
- int[] gbo = NodePool.makeArray();
+ int[] gbo =globalNodepool.makeArray();
gbo[j.getShareOrder()] = 1; // UIMA4275 - only one
j.setGivenByOrder(gbo);
@@ -1585,7 +1588,7 @@ public class NodepoolScheduler
}
logger.info(methodName, j.getId(), "Request is granted a machine for
reservation.");
- int[] gbo = NodePool.makeArray();
+ int[] gbo =globalNodepool.makeArray();
int order = j.getShareOrder(); // memory, coverted to order, so we
can find stuff
gbo[order] = 1;
j.setGivenByOrder(gbo);
@@ -1722,7 +1725,7 @@ public class NodepoolScheduler
logger.info(methodName, null, "Return from", np.getId(),
"proceeding with logic for", nodepool.getId());
}
- int neededByOrder[] = NodePool.makeArray(); // for each order,
how many N-shares do I want to add?
+ int neededByOrder[] =globalNodepool.makeArray(); // for each
order, how many N-shares do I want to add?
// int total_needed = 0;
Map<IRmJob, Integer> overages = new HashMap<IRmJob, Integer>();
// UIMA-4275
@@ -2298,8 +2301,8 @@ public class NodepoolScheduler
NodePool np = allPools[npi];
String id = np.getId();
- int[] vmach = NodePool.makeArray();
- int[] nmach = NodePool.makeArray();
+ int[] vmach =globalNodepool.makeArray();
+ int[] nmach =globalNodepool.makeArray();
Map<IRmJob, Integer> jobmap = new HashMap<IRmJob, Integer>();
vshares.put(id, vmach);
nshares.put(id, nmach);
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ResourceClass.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ResourceClass.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ResourceClass.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/ResourceClass.java
Mon Aug 24 19:25:05 2015
@@ -84,7 +84,7 @@ public class ResourceClass
private static Comparator<IEntity> apportionmentSorter = new
ApportionmentSorterCl();
- public ResourceClass(DuccProperties props, long share_quantum)
+ public ResourceClass(DuccProperties props)
{
//
// We can assume everything useful is here because the parser insured
it
@@ -92,7 +92,7 @@ public class ResourceClass
this.id = props.getStringProperty("name");
this.policy = Policy.valueOf(props.getStringProperty("policy"));
this.priority = props.getIntProperty("priority");
- this.share_quantum = (int) (share_quantum / ( 1024 * 1024 ));
// KB back to GB
+ // (Note: the share quantum is set when the nodepool is set because it
isn't known quite yet in the constructor.)
String userset = props.getProperty("users");
if ( userset != null ) {
@@ -111,7 +111,9 @@ public class ResourceClass
this.enforce_memory = props.getBooleanProperty("enforce", true);
}
- // For now, R 2.0.0 not configurable, and not cappable. Hope to
revive in future release.
+ // This is not used any more - we keep it for back-level
compatibility, and because
+ // we may revive it in the future. It will therefore be referenced,
but by making it
+ // Integer.MAX_VALUE it is essentially a no-op.
this.fair_share_cap = Integer.MAX_VALUE; // UIMA-4275
if ( this.policy == Policy.FAIR_SHARE ) {
@@ -153,9 +155,31 @@ public class ResourceClass
return authorizedUsers.containsKey(user);
}
+ /**
+ * Ask my nodepool to make an array of the right size for the caller.
+ */
+ int[] makeArray()
+ {
+ return nodepool.makeArray();
+ }
+
+ /**
+ * Ask my nodepool what the largest order of job I supoprt is.
+ */
+ int getMaxOrder()
+ {
+ return nodepool.getMaxOrder();
+ }
+
+ public int getShareQuantum()
+ {
+ return share_quantum;
+ }
+
public void setNodepool(NodePool np)
{
this.nodepool = np;
+ this.share_quantum = np.getShareQuantum();
}
public NodePool getNodepool()
@@ -336,6 +360,7 @@ public class ResourceClass
// UIMA-4275
public boolean fairShareCapExceeded(IRmJob j)
{
+ // fair-share caps are deprecated, insure this never returns true
return false;
// if ( policy != Policy.FAIR_SHARE ) return false;
@@ -407,7 +432,7 @@ public class ResourceClass
if ( given_by_order == null ) return; // nothing given, nothing
to adjust
- for ( int o = NodePool.getMaxOrder(); o > 0; o-- ) {
+ for ( int o = nodepool.getMaxOrder(); o > 0; o-- ) {
np.countOutNSharesByOrder(o, given_by_order[o]);
}
}
@@ -505,8 +530,8 @@ public class ResourceClass
public void initWantedByOrder(ResourceClass unused)
{
- int ord = NodePool.getMaxOrder();
- wanted_by_order = NodePool.makeArray();
+ int ord = nodepool.getMaxOrder();
+ wanted_by_order = nodepool.makeArray();
for ( int o = ord; o > 0; o-- ) {
wanted_by_order[o] = countNSharesWanted(o);
wanted_by_order[0] += wanted_by_order[o];
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/RmJob.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/RmJob.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/RmJob.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/RmJob.java
Mon Aug 24 19:25:05 2015
@@ -321,7 +321,7 @@ public class RmJob
public void initWantedByOrder(ResourceClass unused)
{
- wanted_by_order = NodePool.makeArray();
+ wanted_by_order = unused.makeArray();
wanted_by_order[share_order] = getJobCap();
wanted_by_order[0] = wanted_by_order[share_order];
}
@@ -1228,6 +1228,11 @@ public class RmJob
this.threads = th;
}
+ public int getShareQuantum()
+ {
+ return resource_class.getShareQuantum();
+ }
+
public int getMemory() {
return memory;
}
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Scheduler.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Scheduler.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Scheduler.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Scheduler.java
Mon Aug 24 19:25:05 2015
@@ -115,9 +115,6 @@ public class Scheduler
String schedImplName;
IScheduler[] schedulers;
- long share_quantum = 15; // 15 GB - smallest share size -
converted to KB during init because
- // hosts report in KB
-
long share_free_dram = 0; // 0 GB in KB - minim memory
after shares are allocated
long dramOverride = 0; // if > 0, use this instead of
amount reported by agents (modeling and testing)
@@ -174,7 +171,6 @@ public class Scheduler
evictionPolicy = EvictionPolicy.valueOf(ep);
// nodepool = new NodePool(null, evictionPolicy, 0); //
global nodepool
- share_quantum =
SystemPropertyResolver.getLongProperty("ducc.rm.share.quantum", share_quantum)
* 1024 * 1024; // GB -> KB
share_free_dram =
SystemPropertyResolver.getLongProperty("ducc.rm.reserved.dram",
share_free_dram) * 1024 * 1024; // GB -> KB
ducc_home =
SystemPropertyResolver.getStringProperty("DUCC_HOME");
@@ -227,7 +223,6 @@ public class Scheduler
// scheduler.setClasses(resourceClasses);
// scheduler.setNodePool(nodepools[0]);
- logger.info(methodName, null, "Scheduler running with share quantum
: ", (share_quantum / (1024*1024)), " GB");
logger.info(methodName, null, " reserved DRAM
: ", (share_free_dram / (1024*1024)), " GB");
logger.info(methodName, null, " DRAM override
: ", (dramOverride / (1024*1024)), " GB");
logger.info(methodName, null, " scheduler
: ", schedImplName);
@@ -427,10 +422,11 @@ public class Scheduler
/**
* Calculate share order, given some memory size in GB (as in from a job
spec)
*/
- public int calcShareOrder(long mem)
+ public int calcShareOrder(IRmJob j)
{
// Calculate its share order
- mem = mem * 1024 * 1024; // to GB
+ long mem = j.getMemory() << 20 ; // to KB
from GB
+ int share_quantum = j.getShareQuantum();
int share_order = (int) (mem / share_quantum); //
liberal calc, round UP
if ( (mem % share_quantum) > 0 ) {
@@ -439,6 +435,7 @@ public class Scheduler
return share_order;
}
+
/**
* Collect all the classes served by the indicated nodepool (property
set). This fills
* in the 'ret' map from the parameter 'dp' and recursive calls to the
children in dp.
@@ -548,7 +545,7 @@ public class Scheduler
logger.info(methodName, null, ResourceClass.getHeader());
logger.info(methodName, null, ResourceClass.getDashes());
for ( DuccProperties props : cls.values() ) {
- ResourceClass rc = new ResourceClass(props, share_quantum);
+ ResourceClass rc = new ResourceClass(props);
resourceClasses.put(rc, rc);
resourceClassesByName.put(rc.getName(), rc);
logger.info(methodName, null, rc.toString());
@@ -590,7 +587,8 @@ public class Scheduler
@SuppressWarnings("unchecked")
Map<String, String> nodes = (Map<String, String>)
np.get("nodes");
int search_order = np.getIntProperty("search-order", 100);
- nodepools[i] = new NodePool(null, id, nodes, evictionPolicy, 0,
search_order);
+ int q = np.getIntProperty("share-quantum", 15) << 20 ; // to
kB which is how the nodes report in
+ nodepools[i] = new NodePool(null, id, nodes, evictionPolicy, 0,
search_order, q);
schedulers[i].setNodePool(nodepools[i]); // set
its top-level nodepool
mapNodesToNodepool(nodes, nodepools[i]);
@@ -614,6 +612,7 @@ public class Scheduler
}
schedulers[i].setClasses(classesForNp);
+
}
// Here create or update Users with constraints from the registry
@@ -973,7 +972,7 @@ public class Scheduler
j.setUser(u);
// Calculate its share order
- int share_order = calcShareOrder(j.getMemory());
+ int share_order = calcShareOrder(j);
j.setShareOrder(share_order);
// Assign it to its priority class
@@ -1114,7 +1113,7 @@ public class Scheduler
if ( dramOverride > 0 ) {
allocatable_mem = dramOverride;
}
- share_order = (int) (allocatable_mem / share_quantum);
// conservative - rounds down (this will always cast ok)
+ share_order = (int) (allocatable_mem / np.getShareQuantum());
// conservative - rounds down (this will always cast ok)
} else {
share_order = m.getShareOrder();
}
@@ -1229,8 +1228,8 @@ public class Scheduler
ret.setAllMachines(np.countAllLocalMachines());
- int[] onlineMachines = NodePool.makeArray();
- int[] freeMachines = NodePool.makeArray();
+ int[] onlineMachines = np.makeArray();
+ int[] freeMachines = np.makeArray();
for ( int i = 1; i < freeMachines.length; i++ ) {
freeMachines[i] += np.countFreeMachines(i); // (these are
local, as we want)
}
@@ -1272,20 +1271,22 @@ public class Scheduler
break;
}
- int[] demanded = NodePool.makeArray();
- int[] awarded = NodePool.makeArray();
+ // TODO MUST FIX THIS
- HashMap<IRmJob, IRmJob> jobs = cl.getAllJobs();
- for ( IRmJob j : jobs.values() ) {
- int o = j.getShareOrder();
- demanded[o] += j.queryDemand();
- awarded[o] += j.countNShares();
- }
+ // int[] demanded = NodePool.makeArray();
+ // int[] awarded = NodePool.makeArray();
+
+ // HashMap<IRmJob, IRmJob> jobs = cl.getAllJobs();
+ // for ( IRmJob j : jobs.values() ) {
+ // int o = j.getShareOrder();
+ // demanded[o] += j.queryDemand();
+ // awarded[o] += j.countNShares();
+ // }
- qcl.setName(cl.getName());
- qcl.setDemanded(demanded);
- qcl.setAwarded(awarded);
- reply.addClass(qcl);
+ // qcl.setName(cl.getName());
+ // qcl.setDemanded(demanded);
+ // qcl.setAwarded(awarded);
+ // reply.addClass(qcl);
}
}
@@ -1306,8 +1307,6 @@ public class Scheduler
return ret;
}
- ret.setShareQuantum(share_quantum);
-
calculateLoad(ret);
ArrayList<NodePool> allpools = new ArrayList<NodePool>();
@@ -1522,7 +1521,7 @@ public class Scheduler
public void resetNodepools()
{
for ( NodePool np : nodepools ) {
- np.reset(NodePool.getMaxOrder());
+ np.reset(np.getMaxOrder());
}
}
@@ -1555,10 +1554,12 @@ public class Scheduler
{
String methodName = "processRecovery";
- int share_order = calcShareOrder(j.getMemory());
ResourceClass rc = resourceClassesByName.get(j.getClassName());
- j.setShareOrder(share_order);
j.setResourceClass(rc);
+
+ int share_order = calcShareOrder(j);
+ j.setShareOrder(share_order);
+
HashMap<Share, Share> shares = j.getRecoveredShares();
List<Share> sharesToShrink = new ArrayList<Share>(); // UIMA-4142
StringBuffer sharenames = new StringBuffer();
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Share.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Share.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Share.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/Share.java
Mon Aug 24 19:25:05 2015
@@ -168,6 +168,7 @@ public class Share
return machine;
}
+
long getHostMemory()
{
if ( machine != null ) return machine.getMemory();
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/User.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/User.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/User.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-rm/src/main/java/org/apache/uima/ducc/rm/scheduler/User.java
Mon Aug 24 19:25:05 2015
@@ -172,8 +172,8 @@ public class User
public void initWantedByOrder(ResourceClass rc)
{
- wanted_by_order = NodePool.makeArray();
- for ( int o = NodePool.getMaxOrder(); o > 0; o-- ) {
+ wanted_by_order = rc.makeArray();
+ for ( int o = rc.getMaxOrder(); o > 0; o-- ) {
wanted_by_order[o] = countNSharesWanted(o, rc);
wanted_by_order[0] += wanted_by_order[o];
}
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/RmStateDuccEvent.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/RmStateDuccEvent.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/RmStateDuccEvent.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/RmStateDuccEvent.java
Mon Aug 24 19:25:05 2015
@@ -105,9 +105,9 @@ public class RmStateDuccEvent
Map<DuccId, IResource> existing = j.getResources();
if ( existing == null ) {
- buf.append(String.format("%s %s %s\n\tExisting[0]",
j.getDuccType(), j.getId().getFriendly(), reason));
+ buf.append(String.format("%s %s %dGB %s\n\tExisting[0]",
j.getDuccType(), j.getId().getFriendly(), j.memoryGbPerProcess(), reason));
} else {
- buf.append(String.format("%s %s %s\n\tExisting[%d]: ",
j.getDuccType(), j.getId().getFriendly(), reason, existing.size()));
+ buf.append(String.format("%s %s %dGB %s\n\tExisting[%d]: ",
j.getDuccType(), j.getId().getFriendly(), j.memoryGbPerProcess(), reason,
existing.size()));
for ( IResource r : existing.values() ) {
buf.append(r.toString());
buf.append(" ");
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/IRmJobState.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/IRmJobState.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/IRmJobState.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/IRmJobState.java
Mon Aug 24 19:25:05 2015
@@ -77,6 +77,11 @@ public interface IRmJobState extends Ser
boolean isRefused();
/**
+ * Actual memory allocated per process.
+ */
+ public int memoryGbPerProcess();
+
+ /**
* If isRefused() is true, this string contains a message explaining why.
*/
String getReason();
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/Resource.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/Resource.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/Resource.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/Resource.java
Mon Aug 24 19:25:05 2015
@@ -29,7 +29,7 @@ public class Resource implements IResour
private Node node; // Node id, assigned by Agent
private boolean purged; // Purged, for node failure
private int qShares; // Number of quantum shares this
resource occupies
- private transient long itime; // initialization time, for
the toString, but not to be transmitted
+ private transient long itime; // initialization time, for the
toString, but not to be transmitted
// dissallow
@SuppressWarnings("unused")
private Resource()
@@ -42,7 +42,7 @@ public class Resource implements IResour
this.node = node;
this.purged = purged;
this.qShares = qShares;
- this.itime = itime;
+ this.itime = itime;
}
public DuccId getId()
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/RmJobState.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/RmJobState.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/RmJobState.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-transport/src/main/java/org/apache/uima/ducc/transport/event/rm/RmJobState.java
Mon Aug 24 19:25:05 2015
@@ -28,7 +28,9 @@ public class RmJobState implements IRmJo
private static final long serialVersionUID = 1L;
private DuccId duccId; // this job's
DuccId as assigned by OR
- private DuccType ducc_type; // for messages :(
+ private DuccType ducc_type; // for messages
+ private int memoryPerProcess; // in gb, actual
memory allocated, usually > memory requesed
+ // because of rounding
to nearst multiple of the quantum
// for all maps:
// key: DuccId of a share, assigned by RM
@@ -47,11 +49,13 @@ public class RmJobState implements IRmJo
}
public RmJobState(DuccId duccId,
+ int memoryPerProcess,
Map<DuccId, IResource> resources,
Map<DuccId, IResource> removals,
Map<DuccId, IResource> additions)
{
this.duccId = duccId;
+ this.memoryPerProcess = memoryPerProcess;
this.resources = resources;
this.pendingRemovals = removals;
this.pendingAdditions = additions;
@@ -60,6 +64,7 @@ public class RmJobState implements IRmJo
public RmJobState(DuccId duccId, String refusalReason)
{
this.duccId = duccId;
+ this.memoryPerProcess = 0;
this.refused = true;
this.reason = refusalReason;
}
@@ -113,4 +118,16 @@ public class RmJobState implements IRmJo
{
this.ducc_type = dt;
}
+
+ // in GB
+ public int memoryGbPerProcess()
+ {
+ return memoryPerProcess;
+ }
+
+ public int setMemorPerProcess(int m)
+ {
+ return memoryPerProcess;
+ }
+
}
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/JobFragment.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/JobFragment.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/JobFragment.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/JobFragment.java
Mon Aug 24 19:25:05 2015
@@ -33,19 +33,21 @@ class JobFragment
int qshares; // total qshares represented by the node
String service_endpoint; // for services only, the defined endpoint we
we can link back to services page
int mem; // Actual memory requested
+ int quantum; // The scheduling quantum used for this job
String color; // color to draw this
DuccType type; // Job, Service, Reservation, Pop.
String textColor = "white";
String fillColor = "black";
- JobFragment(String user, DuccType type, String id, int mem, int qshares,
String service_endpoint)
+ JobFragment(String user, DuccType type, String id, int mem, int qshares,
int quantum, String service_endpoint)
{
this.user = user;
this.type = type;
this.id = id;
this.qshares = qshares;
this.mem = mem;
+ this.quantum = quantum;
this.nprocesses = 1;
this.service_endpoint = service_endpoint;
setColors();
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/Markup.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/Markup.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/Markup.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/Markup.java
Mon Aug 24 19:25:05 2015
@@ -217,7 +217,7 @@ public class Markup
out.append("<title>");
out.append(j.qshares);
out.append(" unused shares (");
- out.append((j.qshares * NodeViz.quantum));
+ out.append((j.qshares * j.quantum));
out.append("GB) on ");
out.append(h.name);
out.append("(");
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/NodeViz.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/NodeViz.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/NodeViz.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/NodeViz.java
Mon Aug 24 19:25:05 2015
@@ -19,6 +19,8 @@
package org.apache.uima.ducc.ws.server.nodeviz;
+import java.io.FileNotFoundException;
+import java.io.IOException;
import java.util.Arrays;
import java.util.Comparator;
import java.util.HashMap;
@@ -26,8 +28,11 @@ import java.util.Map;
import java.util.concurrent.ConcurrentSkipListMap;
import org.apache.uima.ducc.common.Node;
+import org.apache.uima.ducc.common.NodeConfiguration;
import org.apache.uima.ducc.common.utils.DuccLogger;
import org.apache.uima.ducc.common.utils.DuccLoggerComponents;
+import org.apache.uima.ducc.common.utils.DuccPropertiesResolver;
+import org.apache.uima.ducc.common.utils.IllegalConfigurationException;
import org.apache.uima.ducc.common.utils.SystemPropertyResolver;
import org.apache.uima.ducc.common.utils.Version;
import org.apache.uima.ducc.transport.event.OrchestratorStateDuccEvent;
@@ -61,8 +66,8 @@ public class NodeViz
private String visualization; // cached visualization
private long update_interval = 60000; // Only gen a new viz
every 'this long'
+ static int default_quantum = 4; // for hosts with no
jobs so we don't know or care that much
private String version = "1.1.0";
- static int quantum = 4;
static String wshost = "";
static String wsport = "42133";
static boolean strip_domain = true;
@@ -72,17 +77,16 @@ public class NodeViz
String methodName = "NodeViz";
update_interval =
SystemPropertyResolver.getLongProperty("ducc.viz.update.interval",
update_interval);
- quantum =
SystemPropertyResolver.getIntProperty("ducc.rm.share.quantum", quantum);
-
+ default_quantum =
SystemPropertyResolver.getIntProperty("ducc.rm.share.quantum", default_quantum);
wshost =
SystemPropertyResolver.getStringProperty("ducc.ws.node",
System.getProperty("ducc.head"));
wsport =
SystemPropertyResolver.getStringProperty("ducc.ws.port", wsport);
- strip_domain =
SystemPropertyResolver.getBooleanProperty("ducc.ws.visualization.strip.domain",
true);
+ strip_domain =
SystemPropertyResolver.getBooleanProperty("ducc.ws.visualization.strip.domain",
true);
logger.info(methodName, null,
"------------------------------------------------------------------------------------");
logger.info(methodName, null, "Node Visualization starting:");
logger.info(methodName, null, " DUCC home : ",
System.getProperty("DUCC_HOME"));
logger.info(methodName, null, " ActiveMQ URL : ",
System.getProperty("ducc.broker.url"));
- logger.info(methodName, null, "Using Share Quantum : ",
quantum);
+ logger.info(methodName, null, "Default Quantum : ",
default_quantum);
logger.info(methodName, null, "Viz update Interval : ",
update_interval);
logger.info(methodName, null, "Web Server Host : ",
wshost);
logger.info(methodName, null, "Web Server Port : ",
wsport);
@@ -125,6 +129,33 @@ public class NodeViz
int pop_shares = 0;
int reservation_shares = 0;
+ int job_gb = 0;
+ int service_gb = 0;
+ int pop_gb = 0;
+ int reservation_gb = 0;
+
+ // Must find nost configuration so we can work out the quantum used to
schedule each job
+ String class_definitions = SystemPropertyResolver
+ .getStringProperty(DuccPropertiesResolver
+ .ducc_rm_class_definitions,
"scheduler.classes");
+ String user_registry = SystemPropertyResolver
+ .getStringProperty(DuccPropertiesResolver
+ .ducc_rm_user_registry, "ducc.users");
+ class_definitions = System.getProperty("DUCC_HOME") + "/resources/" +
class_definitions;
+ NodeConfiguration nc = new NodeConfiguration(class_definitions, null,
user_registry, logger); // UIMA-4142 make the config global
+ try {
+ nc.readConfiguration();
+ } catch (FileNotFoundException e1) {
+ // TODO Auto-generated catch block
+ e1.printStackTrace();
+ } catch (IOException e1) {
+ // TODO Auto-generated catch block
+ e1.printStackTrace();
+ } catch (IllegalConfigurationException e1) {
+ // TODO Auto-generated catch block
+ e1.printStackTrace();
+ }
+
// first step, generate the viz from the OR map which seems to have
everything we need
// next stop, walk the machines list and generate empty node for any
machine in that list
// that had no work on it
@@ -157,10 +188,20 @@ public class NodeViz
IDuccStandardInfo si = w.getStandardInfo();
IDuccSchedulingInfo sti = w.getSchedulingInfo();
- String user = si.getUser();
- String duccid = service_id == null ?
Long.toString(w.getDuccId().getFriendly()) : service_id; // UIMA-4209
- int jobmem =
Integer.parseInt(sti.getShareMemorySize());
- int qshares = jobmem / quantum;
+ String user = si.getUser();
+ String duccid = service_id == null ?
Long.toString(w.getDuccId().getFriendly()) : service_id; // UIMA-4209
+ int jobmem =
Integer.parseInt(sti.getShareMemorySize());
+
+ String sclass = sti.getSchedulingClass();
+ int quantum = default_quantum;
+ try {
+ quantum = nc.getShareQuantum(sclass);
+ } catch ( Exception e ) {
+ // this most likely caused by a reconfigure so that a job's
class no longer exists. nothing to do about it
+ // but punt and try not to crash.
+ logger.warn(methodName, null, "Cannot find scheduling class or
quantum for " + sclass + ". Using default quantum of " + default_quantum);
+ }
+ int qshares = jobmem / quantum;
if ( jobmem % quantum != 0 ) qshares++;
switch ( type ) {
@@ -185,12 +226,15 @@ public class NodeViz
switch ( type ) {
case Job:
job_shares += qshares;
+ job_gb += jobmem;
break;
case Pop:
pop_shares += qshares;
+ pop_gb += jobmem;
break;
case Service:
service_shares += qshares;
+ service_gb += jobmem;
break;
}
@@ -216,6 +260,7 @@ public class NodeViz
logger.debug(methodName, w.getDuccId(), "Receive:", type,
w.getStateObject(), "processes[", rm.size(), "] Completed:", w.isCompleted());
reservation_shares += qshares;
+ reservation_gb += jobmem;
for ( IDuccReservation r: rm.values()) {
Node n = r.getNode();
@@ -244,9 +289,11 @@ public class NodeViz
logger.debug(methodName, null, "Generateing visualizaiton");
ConcurrentSkipListMap<String,MachineInfo> m =
machineData.getMachines();
-
for (String s : m.keySet()) {
-
+ //
+ // This is for hosts that have no work on them so they didn't come
in the work map
+ //
+
MachineInfo mi = m.get(s);
// NOTE: the map changes all the time so the value may be gone.
This situation
// will be fixed one day but for now just forget the node,
it will show up
@@ -258,24 +305,24 @@ public class NodeViz
String key = strip(s); // our key, possibly with
domain stripped
if ( ! hosts.containsKey(key) ) {
// System.out.println("Set host from MachineInfo with key :" +
key + ":");
- VisualizedHost vh = new VisualizedHost(mi, quantum);
+
+ VisualizedHost vh = new VisualizedHost(mi, default_quantum);
hosts.put(key, vh);
}
}
- int total_shares = 0;
- int total_ram = 0;
+ int total_gb = 0;
Markup markup = new Markup();
VisualizedHost[] sorted = hosts.values().toArray(new
VisualizedHost[hosts.size()]);
Arrays.sort(sorted, new HostSorter());
for ( VisualizedHost vh : sorted ) {
vh.toSvg(markup);
- total_shares += vh.countShares();
- total_ram += vh.countRam();
+ total_gb += vh.countRam();
}
String page = markup.close();
- int unoccupied_shares = total_shares - (job_shares + pop_shares +
service_shares + reservation_shares);
+ int unoccupied_gb = total_gb - (job_gb + pop_gb + service_gb +
reservation_gb);
+ int total_shares = job_shares + pop_shares + service_shares +
reservation_shares;
visualization =
"<html>" +
@@ -285,20 +332,19 @@ public class NodeViz
"<i onclick=\"ducc_viz_node_sorter('size')\"
id=\"ducc-viz-sort-size\" style=\"color:red\">Size </i>" +
"<i onclick=\"ducc_viz_node_sorter('name')\"
id=\"ducc-viz-sort-name\"\">Name</i>" +
"</br>" +
- "<b>Shares of size " + quantum + "GB: </b>" + total_shares +
+ "<b>Total shares: </b>" + total_shares +
", <b>Jobs: </b>" + job_shares +
", <b>Services: </b>" + service_shares +
", <b>Managed Reservations: </b>" + pop_shares +
", <b>Reservations: </b>" + reservation_shares +
- ", <b>Unoccupied: </b>" + unoccupied_shares +
+ ", <b>Unoccupied: </b>" + unoccupied_gb +
"<br><i><small>" +
- "<b>RAM Total:</b> " + total_ram +
- "GB, <b>For shares:</b> " + (total_shares * quantum) +
- "GB, <b>Jobs:</b> " + (job_shares * quantum) +
- "GB, <b>Services:</b> " + (service_shares * quantum) +
- "GB, <b>Managed Reservations:</b> " + (pop_shares * quantum) +
- "GB, <b>Reservations:</b> " + (reservation_shares * quantum) +
- "GB, <b>Unoccupied:</b> " + (unoccupied_shares * quantum) +
+ "<b>RAM Total:</b> " + total_gb +
+ "GB, <b>Jobs:</b> " + (job_gb) +
+ "GB, <b>Services:</b> " + (service_gb) +
+ "GB, <b>Managed Reservations:</b> " + (pop_gb) +
+ "GB, <b>Reservations:</b> " + (pop_gb) +
+ "GB, <b>Unoccupied:</b> " + (unoccupied_gb) +
"GB</small></i>" +
"</div>" +
"<br>" +
Modified:
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/VisualizedHost.java
URL:
http://svn.apache.org/viewvc/uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/VisualizedHost.java?rev=1697485&r1=1697484&r2=1697485&view=diff
==============================================================================
---
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/VisualizedHost.java
(original)
+++
uima/sandbox/uima-ducc/trunk/uima-ducc-web/src/main/java/org/apache/uima/ducc/ws/server/nodeviz/VisualizedHost.java
Mon Aug 24 19:25:05 2015
@@ -132,7 +132,7 @@ class VisualizedHost
}
if ( ! found ) {
logger.debug(methodName, null, name, "Create new job fragment
for", user, "with", qshares, "qshare, type", type);
- JobFragment j = new JobFragment(user, type, duccid, jobmem,
qshares, service_endpoint);
+ JobFragment j = new JobFragment(user, type, duccid, jobmem,
qshares, quantum, service_endpoint);
fragments.add(j);
}
}
@@ -166,7 +166,7 @@ class VisualizedHost
m.tooltipEnd();
Collections.sort(fragments, sorter);
- float height_one_share = (float) Math.sqrt(shares * NodeViz.quantum) /
shares;
+ float height_one_share = (float) Math.sqrt(shares * quantum) / shares;
float foo = (float) Math.sqrt(mem) / shares;
logger.debug(methodName, null, name, "shares", shares,
"height-one-share", height_one_share, "foo", foo);
float top = 0f + TITLE_ADJUSTMENT; // the top of the
box
