[cuebot/cuegui/pycue/rqd] Store PSS and MaxPSS

VERSION.in

@@ -1 +1 @@
-1.15
+1.16

cuebot/src/main/java/com/imageworks/spcue/FrameDetail.java

@@ -27,6 +27,7 @@ public class FrameDetail extends FrameEntity implements FrameInterface {
     public int retryCount;
     public int exitStatus;
     public long maxRss;
+    public long maxPss;
     public int dispatchOrder;
     public String lastResource;
 

cuebot/src/main/java/com/imageworks/spcue/dao/FrameDao.java

@@ -326,10 +326,13 @@ public interface FrameDao {
      * @param f
      * @param maxRss
      * @param rss
+     * @param maxPss
+     * @param pss
      * @param lluTime
      * @throws FrameReservationException if the frame is locked by another thread.
      */
-    void updateFrameMemoryUsageAndLluTime(FrameInterface f, long maxRss, long rss, long lluTime);
+    void updateFrameMemoryUsageAndLluTime(FrameInterface f, long maxRss, long rss, long maxPss,
+            long pss, long lluTime);
 
     /**
      * Attempt to put a exclusive row lock on the given frame. The frame must be in the specified

cuebot/src/main/java/com/imageworks/spcue/dao/JobDao.java

@@ -365,6 +365,15 @@ public interface JobDao {
      */
     void updateMaxRSS(JobInterface job, long maxRss);
 
+    /**
+     * Update jobs max PSS. Only updates if the passed in value is greater than the current value of
+     * int_max_pss
+     *
+     * @param job
+     * @param maxPss
+     */
+    void updateMaxPSS(JobInterface job, long maxPss);
+
     /**
      * Inserts a key/value pair into the jobs env table
      *

cuebot/src/main/java/com/imageworks/spcue/dao/LayerDao.java

@@ -195,6 +195,16 @@ public interface LayerDao {
      */
     void updateLayerMaxRSS(LayerInterface layer, long val, boolean force);
 
+    /**
+     * Updates the layer's maximum PSS value. If force is true, the max RSS is updated no matter
+     * what the current value is. If force is false, the value is only updated the val is greater
+     * than than the existing value.
+     *
+     * @param layer
+     * @param val
+     */
+    void updateLayerMaxPSS(LayerInterface layer, long val, boolean force);
+
     /**
      * Increases the value of the minimum memory when the supplied value is larger than the current
      * value

cuebot/src/main/java/com/imageworks/spcue/dao/ProcDao.java

@@ -133,8 +133,9 @@ public interface ProcDao {
      * @param usedKb
      * @param maxKb
      */
-    void updateProcMemoryUsage(FrameInterface f, long rss, long maxRss, long vsize, long maxVsize,
-            long usedGpuMemory, long maxUsedGpuMemory, long usedSwapMemory, byte[] children);
+    void updateProcMemoryUsage(FrameInterface f, long rss, long maxRss, long pss, long maxPss,
+            long vsize, long maxVsize, long usedGpuMemory, long maxUsedGpuMemory,
+            long usedSwapMemory, byte[] children);
 
     /**
      * get aq virual proc from its unique id

cuebot/src/main/java/com/imageworks/spcue/dao/postgres/FrameDaoJdbc.java

@@ -293,6 +293,7 @@ public FrameDetail mapRow(ResultSet rs, int rowNum) throws SQLException {
             frame.layerId = rs.getString("pk_layer");
             frame.showId = rs.getString("pk_show");
             frame.maxRss = rs.getLong("int_mem_max_used");
+            frame.maxPss = rs.getLong("int_pss_max_used");
             frame.name = rs.getString("str_name");
             frame.number = rs.getInt("int_number");
             frame.dispatchOrder = rs.getInt("int_dispatch_order");
@@ -709,14 +710,15 @@ public ResourceUsage getResourceUsage(FrameInterface f) {
                 RESOURCE_USAGE_MAPPER, f.getFrameId());
     }
 
-    private static final String UPDATE_FRAME_MEMORY_USAGE_AND_LLU_TIME = "UPDATE " + "frame "
-            + "SET " + "ts_updated = current_timestamp," + "int_mem_max_used = ?,"
-            + "int_mem_used = ?," + "ts_llu = ? " + "WHERE " + "pk_frame = ? ";
+    private static final String UPDATE_FRAME_MEMORY_USAGE_AND_LLU_TIME =
+            "UPDATE " + "frame " + "SET " + "ts_updated = current_timestamp,"
+                    + "int_mem_max_used = ?," + "int_mem_used = ?," + "int_pss_max_used = ?,"
+                    + "int_pss_used = ?," + "ts_llu = ? " + "WHERE " + "pk_frame = ? ";
 
     @Override
     public void updateFrameMemoryUsageAndLluTime(FrameInterface f, long maxRss, long rss,
-            long lluTime) {
-        getJdbcTemplate().update(UPDATE_FRAME_MEMORY_USAGE_AND_LLU_TIME, maxRss, rss,
+            long maxPss, long pss, long lluTime) {
+        getJdbcTemplate().update(UPDATE_FRAME_MEMORY_USAGE_AND_LLU_TIME, maxRss, rss, maxPss, pss,
                 new Timestamp(lluTime * 1000l), f.getFrameId());
     }
 

cuebot/src/main/java/com/imageworks/spcue/dao/postgres/JobDaoJdbc.java

@@ -361,6 +361,12 @@ public void updateMaxRSS(JobInterface job, long value) {
                 job.getJobId(), value);
     }
 
+    public void updateMaxPSS(JobInterface job, long value) {
+        getJdbcTemplate().update(
+                "UPDATE job_mem SET int_max_pss=? WHERE pk_job=? AND int_max_pss < ?", value,
+                job.getJobId(), value);
+    }
+
     private static final String UPDATE_JOB_FINISHED = "UPDATE " + "job " + "SET "
             + "str_state = ?, " + "str_visible_name = NULL, " + "ts_stopped = current_timestamp "
             + "WHERE " + "str_state = 'PENDING' " + "AND " + "pk_job = ?";

cuebot/src/main/java/com/imageworks/spcue/dao/postgres/LayerDaoJdbc.java

@@ -357,6 +357,22 @@ public void updateLayerMaxRSS(LayerInterface layer, long val, boolean force) {
         getJdbcTemplate().update(sb.toString(), options);
     }
 
+    private static final String UPDATE_LAYER_MAX_PSS =
+            "UPDATE " + "layer_mem " + "SET " + "int_max_pss = ? " + "WHERE " + "pk_layer = ?";
+
+    @Override
+    public void updateLayerMaxPSS(LayerInterface layer, long val, boolean force) {
+        StringBuilder sb = new StringBuilder(UPDATE_LAYER_MAX_PSS);
+        Object[] options;
+        if (!force) {
+            options = new Object[] {val, layer.getLayerId(), val};
+            sb.append(" AND int_max_pss < ?");
+        } else {
+            options = new Object[] {val, layer.getLayerId()};
+        }
+        getJdbcTemplate().update(sb.toString(), options);
+    }
+
     @Override
     public void updateLayerTags(LayerInterface layer, Set<String> tags) {
         if (tags.size() == 0) {

cuebot/src/main/java/com/imageworks/spcue/dao/postgres/NestedWhiteboardDaoJdbc.java

@@ -86,8 +86,8 @@ public CachedJobWhiteboardMapper(NestedJobWhiteboardMapper result) {
             + "(job_resource.int_gpus + job_resource.int_local_gpus) AS int_gpus, "
             + "job_resource.int_min_cores, " + "job_resource.int_min_gpus, "
             + "job_resource.int_max_cores, " + "job_resource.int_max_gpus, "
-            + "job_mem.int_max_rss " + "FROM " + "show, " + "dept, " + "folder_level, "
-            + "folder_resource, " + "folder " + "LEFT JOIN " + "job " + "ON "
+            + "job_mem.int_max_rss, " + "job_mem.int_max_pss " + "FROM " + "show, " + "dept, "
+            + "folder_level, " + "folder_resource, " + "folder " + "LEFT JOIN " + "job " + "ON "
             + " (folder.pk_folder = job.pk_folder AND job.str_state='PENDING') " + "LEFT JOIN "
             + "facility " + "ON " + "(job.pk_facility = facility.pk_facility) " + "LEFT JOIN "
             + "job_stat " + "ON " + "(job.pk_job = job_stat.pk_job) " + "LEFT JOIN "
@@ -290,6 +290,7 @@ private static final NestedJob mapResultSetToJob(ResultSet rs) throws SQLExcepti
             + "host_stat.int_load, " + "proc.pk_proc, " + "proc.int_cores_reserved AS proc_cores, "
             + "proc.int_gpus_reserved AS proc_gpus, " + "proc.int_mem_reserved AS proc_memory, "
             + "proc.int_mem_used AS used_memory, " + "proc.int_mem_max_used AS max_memory, "
+            + "proc.int_pss_used AS used_pss, " + "proc.int_pss_max_used AS max_pss, "
             + "proc.int_gpu_mem_reserved AS proc_gpu_memory, " + "proc.ts_ping, "
             + "proc.ts_booked, " + "proc.ts_dispatched, " + "proc.b_unbooked, "
             + "redirect.str_name AS str_redirect, " + "job.str_name AS job_name, "
@@ -390,6 +391,8 @@ public NestedHost mapRow(ResultSet rs, int row) throws SQLException {
                                     .setReservedMemory(rs.getLong("proc_memory"))
                                     .setReservedGpuMemory(rs.getLong("proc_gpu_memory"))
                                     .setUsedMemory(rs.getLong("used_memory"))
+                                    .setUsedPss(rs.getLong("used_pss"))
+                                    .setMaxPss(rs.getLong("max_pss"))
                                     .setFrameName(rs.getString("frame_name"))
                                     .setJobName(rs.getString("job_name"))
                                     .setShowName(rs.getString("show_name"))

cuebot/src/main/java/com/imageworks/spcue/dao/postgres/ProcDaoJdbc.java

@@ -171,16 +171,16 @@ public boolean clearVirtualProcAssignment(FrameInterface frame) {
                 frame.getFrameId()) == 1;
     }
 
-    private static final String UPDATE_PROC_MEMORY_USAGE =
-            "UPDATE " + "proc " + "SET " + "int_mem_used = ?, " + "int_mem_max_used = ?,"
-                    + "int_virt_used = ?, " + "int_virt_max_used = ?, " + "int_gpu_mem_used = ?, "
-                    + "int_gpu_mem_max_used = ?, " + "int_swap_used = ?, " + "bytea_children = ?, "
-                    + "ts_ping = current_timestamp " + "WHERE " + "pk_frame = ?";
+    private static final String UPDATE_PROC_MEMORY_USAGE = "UPDATE " + "proc " + "SET "
+            + "int_mem_used = ?, " + "int_mem_max_used = ?," + "int_pss_used = ?, "
+            + "int_pss_max_used = ?, " + "int_virt_used = ?, " + "int_virt_max_used = ?, "
+            + "int_gpu_mem_used = ?, " + "int_gpu_mem_max_used = ?, " + "int_swap_used = ?, "
+            + "bytea_children = ?, " + "ts_ping = current_timestamp " + "WHERE " + "pk_frame = ?";
 
     @Override
-    public void updateProcMemoryUsage(FrameInterface f, long rss, long maxRss, long vss,
-            long maxVss, long usedGpuMemory, long maxUsedGpuMemory, long usedSwapMemory,
-            byte[] children) {
+    public void updateProcMemoryUsage(FrameInterface f, long rss, long maxRss, long pss,
+            long maxPss, long vss, long maxVss, long usedGpuMemory, long maxUsedGpuMemory,
+            long usedSwapMemory, byte[] children) {
         /*
          * This method is going to repeat for a proc every 1 minute, so if the proc is being touched
          * by another thread, then return quietly without updating memory usage.
@@ -202,13 +202,15 @@ public PreparedStatement createPreparedStatement(Connection conn)
                                 conn.prepareStatement(UPDATE_PROC_MEMORY_USAGE);
                         updateProc.setLong(1, rss);
                         updateProc.setLong(2, maxRss);
-                        updateProc.setLong(3, vss);
-                        updateProc.setLong(4, maxVss);
-                        updateProc.setLong(5, usedGpuMemory);
-                        updateProc.setLong(6, maxUsedGpuMemory);
-                        updateProc.setLong(7, usedSwapMemory);
-                        updateProc.setBytes(8, children);
-                        updateProc.setString(9, f.getFrameId());
+                        updateProc.setLong(3, pss);
+                        updateProc.setLong(4, maxPss);
+                        updateProc.setLong(5, vss);
+                        updateProc.setLong(6, maxVss);
+                        updateProc.setLong(7, usedGpuMemory);
+                        updateProc.setLong(8, maxUsedGpuMemory);
+                        updateProc.setLong(9, usedSwapMemory);
+                        updateProc.setBytes(10, children);
+                        updateProc.setString(11, f.getFrameId());
                         return updateProc;
                     }
                 });

cuebot/src/main/java/com/imageworks/spcue/dao/postgres/WhiteboardDaoJdbc.java

@@ -1124,7 +1124,7 @@ public static JobStats mapJobStats(ResultSet rs) throws SQLException {
         JobStats.Builder statsBuilder = JobStats.newBuilder()
                 .setReservedCores(Convert.coreUnitsToCores(rs.getInt("int_cores")))
                 .setReservedGpus(rs.getInt("int_gpus")).setMaxRss(rs.getLong("int_max_rss"))
-                .setTotalFrames(rs.getInt("int_frame_count"))
+                .setMaxPss(rs.getLong("int_max_pss")).setTotalFrames(rs.getInt("int_frame_count"))
                 .setTotalLayers(rs.getInt("int_layer_count"))
                 .setWaitingFrames(rs.getInt("int_waiting_count"))
                 .setRunningFrames(rs.getInt("int_running_count"))
@@ -1188,6 +1188,7 @@ public Layer mapRow(ResultSet rs, int rowNum) throws SQLException {
             LayerStats.Builder statsBuilder = LayerStats.newBuilder()
                     .setReservedCores(Convert.coreUnitsToCores(rs.getInt("int_cores")))
                     .setReservedGpus(rs.getInt("int_gpus")).setMaxRss(rs.getLong("int_max_rss"))
+                    .setMaxPss(rs.getLong("int_max_pss"))
                     .setTotalFrames(rs.getInt("int_total_count"))
                     .setWaitingFrames(rs.getInt("int_waiting_count"))
                     .setRunningFrames(rs.getInt("int_running_count"))
@@ -1309,8 +1310,10 @@ public Frame mapRow(ResultSet rs, int rowNum) throws SQLException {
                     .setState(FrameState.valueOf(SqlUtil.getString(rs, "str_state")))
                     .setLayerName(SqlUtil.getString(rs, "layer_name"))
                     .setUsedMemory(rs.getLong("int_mem_used"))
+                    .setUsedPss(rs.getLong("int_pss_used"))
                     .setReservedMemory(rs.getLong("int_mem_reserved"))
                     .setReservedGpuMemory(rs.getLong("int_gpu_mem_reserved"))
+                    .setMaxPss(rs.getLong("int_pss_max_used"))
                     .setCheckpointState(
                             CheckpointState.valueOf(SqlUtil.getString(rs, "str_checkpoint_state")))
                     .setCheckpointCount(rs.getInt("int_checkpoint_count"));
@@ -1454,7 +1457,8 @@ public Show mapRow(ResultSet rs, int rowNum) throws SQLException {
             + "frame.int_dispatch_order," + "frame.ts_started," + "frame.ts_stopped,"
             + "frame.ts_llu," + "frame.int_retries," + "frame.str_state," + "frame.str_host,"
             + "frame.int_cores," + "frame.int_gpus," + "frame.int_mem_max_used,"
-            + "frame.int_mem_used, " + "frame.int_mem_reserved, " + "frame.int_gpu_mem_reserved, "
+            + "frame.int_mem_used, " + "frame.int_pss_max_used," + "frame.int_pss_used, "
+            + "frame.int_mem_reserved, " + "frame.int_gpu_mem_reserved, "
             + "frame.str_checkpoint_state," + "frame.int_checkpoint_count,"
             + "frame.int_total_past_core_time," + "frame.int_total_past_gpu_time,"
             + "layer.str_name AS layer_name," + "job.str_name AS job_name,"
@@ -1608,7 +1612,7 @@ public Show mapRow(ResultSet rs, int rowNum) throws SQLException {
             + "job_usage.int_core_time_fail, " + "job_usage.int_gpu_time_success, "
             + "job_usage.int_gpu_time_fail, " + "job_usage.int_frame_success_count, "
             + "job_usage.int_frame_fail_count, " + "job_usage.int_clock_time_high,"
-            + "job_usage.int_clock_time_success," + "job_mem.int_max_rss,"
+            + "job_usage.int_clock_time_success," + "job_mem.int_max_rss," + "job_mem.int_max_pss,"
             + "(job_resource.int_cores + job_resource.int_local_cores) AS int_cores,"
             + "(job_resource.int_gpus + job_resource.int_local_gpus) AS int_gpus, "
             + "job.str_loki_url " + "FROM " + "job," + "folder," + "show," + "facility,"
@@ -1627,11 +1631,11 @@ public Show mapRow(ResultSet rs, int rowNum) throws SQLException {
             + "layer_usage.int_frame_success_count, " + "layer_usage.int_frame_fail_count, "
             + "layer_usage.int_clock_time_low, " + "layer_usage.int_clock_time_high,"
             + "layer_usage.int_clock_time_success," + "layer_usage.int_clock_time_fail,"
-            + "layer_mem.int_max_rss," + "layer_resource.int_cores," + "layer_resource.int_gpus "
-            + "FROM " + "layer, " + "job," + "layer_stat, " + "layer_resource, " + "layer_usage, "
-            + "layer_mem " + "WHERE " + "layer.pk_job = job.pk_job " + "AND "
-            + "layer.pk_layer = layer_stat.pk_layer " + "AND "
-            + "layer.pk_layer = layer_resource.pk_layer " + "AND "
+            + "layer_mem.int_max_rss," + "layer_mem.int_max_pss," + "layer_resource.int_cores,"
+            + "layer_resource.int_gpus " + "FROM " + "layer, " + "job," + "layer_stat, "
+            + "layer_resource, " + "layer_usage, " + "layer_mem " + "WHERE "
+            + "layer.pk_job = job.pk_job " + "AND " + "layer.pk_layer = layer_stat.pk_layer "
+            + "AND " + "layer.pk_layer = layer_resource.pk_layer " + "AND "
             + "layer.pk_layer = layer_usage.pk_layer " + "AND "
             + "layer.pk_layer = layer_mem.pk_layer";
 
@@ -1645,7 +1649,7 @@ public Show mapRow(ResultSet rs, int rowNum) throws SQLException {
             + "layer_usage.int_frame_fail_count, " + "layer_usage.int_clock_time_low, "
             + "layer_usage.int_clock_time_high, " + "layer_usage.int_clock_time_success, "
             + "layer_usage.int_clock_time_fail, " + "layer_mem.int_max_rss, "
-            + "layer_resource.int_cores, " + "layer_resource.int_gpus, "
+            + "layer_mem.int_max_pss, " + "layer_resource.int_cores, " + "layer_resource.int_gpus, "
             + "limit_names.str_limit_names " + "FROM " + "layer " + "JOIN "
             + "job ON layer.pk_job = job.pk_job " + "JOIN "
             + "layer_stat ON layer.pk_layer = layer_stat.pk_layer " + "JOIN "
@@ -1746,8 +1750,8 @@ public Show mapRow(ResultSet rs, int rowNum) throws SQLException {
                     + "frame.int_number," + "frame.int_dispatch_order," + "frame.ts_started,"
                     + "frame.ts_stopped," + "frame.ts_llu," + "frame.int_retries,"
                     + "frame.str_state," + "frame.str_host," + "frame.int_cores,"
-                    + "frame.int_mem_max_used," + "frame.int_mem_used, "
-                    + "frame.int_mem_reserved, " + "frame.int_gpus,"
+                    + "frame.int_mem_max_used," + "frame.int_mem_used, " + "frame.int_pss_max_used,"
+                    + "frame.int_pss_used, " + "frame.int_mem_reserved, " + "frame.int_gpus,"
                     + "frame.int_gpu_mem_max_used, " + "frame.int_gpu_mem_used, "
                     + "frame.int_gpu_mem_reserved, " + "frame.str_checkpoint_state,"
                     + "frame.int_checkpoint_count," + "frame.int_total_past_core_time,"

cuebot/src/main/java/com/imageworks/spcue/dispatcher/DispatchSupport.java

@@ -389,10 +389,12 @@ public interface DispatchSupport {
      * @param frame
      * @param rss
      * @param maxRss
+     * @param pss
+     * @param maxPss
      * @param lluTime
      */
-    void updateFrameMemoryUsageAndLluTime(FrameInterface frame, long rss, long maxRss,
-            long lluTime);
+    void updateFrameMemoryUsageAndLluTime(FrameInterface frame, long rss, long maxRss, long pss,
+            long maxPss, long lluTime);
 
     /**
      * Update memory usage data for a given frame's proc record. The frame is used to update the
@@ -401,14 +403,16 @@ void updateFrameMemoryUsageAndLluTime(FrameInterface frame, long rss, long maxRs
      * @param frame
      * @param rss
      * @param maxRss
+     * @param pss
+     * @param maxPss
      * @param vsize
      * @param maxVsize
      * @param usedGpuMemory
      * @param maxUsedGpuMemory
      */
-    void updateProcMemoryUsage(FrameInterface frame, long rss, long maxRss, long vsize,
-            long maxVsize, long usedGpuMemory, long maxUsedGpuMemory, long usedSwapMemory,
-            byte[] children);
+    void updateProcMemoryUsage(FrameInterface frame, long rss, long maxRss, long pss, long maxPss,
+            long vsize, long maxVsize, long usedGpuMemory, long maxUsedGpuMemory,
+            long usedSwapMemory, byte[] children);
 
     /**
      * Return true if adding the given core units would put the show over its burst value.