statistics/node_history_ttr.r

   1 source("functions.r");
   2
   3 nsh <- read.csv('node_status_history.csv', sep=',', header=TRUE)
   4
   5 # system("./harvest_nodehistory.py > node_status_history_nopcu.csv")
   6 nsh_nopcu <- read.csv('node_status_history_nopcu.csv', sep=',', header=TRUE)
   7
   8 nsh_m1 <- read.csv('node_status_history_m1.csv', sep=',', header=TRUE)
   9 # system("stats-m1/harvest_nodehistory_m1.py > ./node_status_history_m1_nopcu.csv")
  10 nsh_m1_nopcu <- read.csv('node_status_history_m1_nopcu.csv', sep=',', header=TRUE)
  11 nsh_m1_nopcu_may <- read.csv('node_status_history_m1_nopcu_may08sep08.csv', sep=',', header=TRUE)
  12
  13 node_hist_image <- function (t, year, from, to, max=0, type="week", title="")
  14 {
  15     dates <-seq(as.Date(from), as.Date(to), type)
  16     months <- format(dates, "%b-%d")
  17     hbreaks<-unclass(as.POSIXct(dates))
  18
  19     image <- matrix(data=0, nrow=max(as.numeric(t$hostname)), ncol=length(hbreaks))
  20
  21     for ( i in seq(1, length(hbreaks)) )
  22     {
  23         # find the range : d plus a day
  24         d <- hbreaks[i]
  25         d_end <- d+60*60*24
  26         # find unique hosts in this day range
  27         t_sub <- t[which(t$start > d & t$start <= d_end & t$status == 'down'),]
  28         unique_hosts <- unique(t_sub$hostname)
  29         if (length(unique_hosts) == 0 ) { next }
  30
  31         host_n_list <- unique_hosts
  32         host_s_list <- as.character(unique_hosts)
  33
  34         for ( hi in seq(1, length(unique_hosts))  )
  35         {
  36             host_s <- host_s_list[hi]
  37             host_n <- host_n_list[hi]
  38             # events for this host after d (to avoid already identified events)
  39             ev <- t[which(t$hostname == host_s & t$start > d ),]
  40             print (sprintf("events length for host %s %s", host_s, length(ev$start)));
  41             # get down events for this host
  42             down_ev_index_list <- which(ev$status == 'down')
  43             for ( e_i in down_ev_index_list )
  44             {
  45                 if ( e_i == length(ev$status) ) {
  46                     # then the node ends down, so fill in the rest with 1.
  47                     for ( j in seq(i,length(hbreaks)) ) {
  48                         image[host_n,j] <- 1
  49                     }
  50                 } else {
  51                     # then there is a subsequent 'good' event
  52                     good_ev <- ev[e_i+1,]
  53                     down_ev <- ev[e_i,]
  54                     dbreaks <- seq(d,good_ev$start+60*60*24,60*60*24)
  55                     # for every index for time d, to good_ev$start
  56                     l<-length(dbreaks)
  57                     print (sprintf("length %s",l));
  58                     for ( j in seq(i,i+l) )
  59                     {
  60                         image[host_n,j] <- 1
  61                     }
  62                 }
  63             }
  64         }
  65     }
  66     myImagePlot(image, xLabels=months, yLabels=c(""), title=title)
  67     return (image);
  68 }
  69
  70
  71
  72 node_hist_dist <- function (t, year, from, to, max=0, type="week", title="")
  73 {
  74     dates <-seq(as.Date(from), as.Date(to), type)
  75     months <- format(dates, "%b-%d")
  76     hbreaks<-unclass(as.POSIXct(dates))
  77     current_ts <- unclass(as.POSIXct(Sys.Date()))
  78
  79     dist <- NULL
  80
  81     unique_hosts <- unique(t$hostname)
  82     host_n_list <- unique_hosts
  83     host_s_list <- as.character(unique_hosts)
  84
  85     down <- 0
  86
  87     for ( hi in seq(1, length(unique_hosts))  )
  88     {
  89         host_s <- host_s_list[hi]
  90         host_n <- host_n_list[hi]
  91         # events for this host after d (to avoid already identified events)
  92         ev <- t[which( t$hostname == host_s ),]
  93         print (sprintf("events length for host %s %s", host_s, length(ev$start)));
  94         # get down events for this host
  95         down_ev_index_list <- which(ev$status == 'down')
  96         for ( e_i in down_ev_index_list )
  97         {
  98             # when equal, there is no resolution so leave it as down
  99             if ( e_i != length(ev$status) ) {
 100                 good_ev <- ev[e_i+1,]
 101                 down_ev <- ev[e_i,]
 102                 dist <- c(dist, good_ev$start - down_ev$start)
 103             } else if ( e_i == length(ev$status) && length(ev$status) == 1) {
 104                 print (sprintf("DOWN FOREVER! %s", length(ev$start) ))
 105                 down <- down + 1
 106                 dist <- c(dist, 10*current_ts - ev$start)
 107             }
 108         }
 109     }
 110     print(down);
 111     return (dist);
 112 }
 113
 114 nsh_image <- node_hist_image(nsh, '2009', '2009-06-01', '2010-02-28', 0, 'day')
 115 nsh_image_m1 <- node_hist_image(nsh_m1, '2009', '2008-10-01', '2009-03-28', 0, 'day')
 116
 117
 118 nsh_short <- nsh[which(nsh$start > unclass(as.POSIXct("2009-06-01", origin="1970-01-01"))[1]),]
 119 nsh_short <- nsh_short[which(nsh_short$start < unclass(as.POSIXct("2009-10-31", origin="1970-01-01"))[1]),]
 120
 121
 122 nsh_short <- nsh_nopcu
 123 nsh_dist <- node_hist_dist(nsh_short, '2009', '2009-06-01', '2010-02-28', 0, 'day')
 124 d<- ecdf(nsh_dist/(60*60*24))
 125
 126 #nsh_m1_short <- nsh_m1[which(nsh_m1$start > unclass(as.POSIXct("2008-10-01", origin="1970-01-01"))[1]),]
 127 nsh_m1_short <- nsh_m1_nopcu
 128 # NOTE: something happened betweeen 10-2 and 10-3
 129 t_1015 <- unclass(as.POSIXct("2008-10-15", origin="1970-01-01"))[1]
 130 t_0224 <- unclass(as.POSIXct("2009-02-24", origin="1970-01-01"))[1]
 131 nsh_m1_short <- nsh_m1_nopcu[which(nsh_m1_nopcu$start > t_1015 & nsh_m1_nopcu$start <= t_0224),]
 132
 133 nsh_m1_short <- nsh_m1_nopcu
 134 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-10-01', '2009-03-22', 0, 'day')
 135 d_m1<- ecdf(nsh_dist_m1/(60*60*24))
 136
 137
 138 t_0530 <- unclass(as.POSIXct("2008-05-30", origin="1970-01-01"))[1]
 139 t_0815 <- unclass(as.POSIXct("2008-08-15", origin="1970-01-01"))[1]
 140 nsh_m1_short <- nsh_m1_nopcu_may[which(nsh_m1_nopcu_may$start > t_0530 & nsh_m1_nopcu_may$start <= t_0815),]
 141 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-05-10', '2008-08-15', 0, 'day')
 142 d_m1_may <- ecdf(nsh_dist_m1/(60*60*24))
 143
 144
 145 # d<-ecdf(nsh_dist[which(nsh_dist/(60*60*24) < 90 )]/(60*60*24)),
 146 # 180 ~= 6 months.
 147 par(mfrow=c(1,1))
 148 par(mai=c(.9,.9,.1,.1))
 149 start_image("node_history_ttr_nopcu.png")
 150 plot(d, xlim=c(0,180), ylim=c(0,1), axes=F, xlab="Days to Resolve", ylab="Percentile",
 151    col.hor='red', col.vert='red', pch='.', col.points='red', main="")
 152 plot(d_m1, xlim=c(0,180), ylim=c(0,1), xlab="Days to Resolve", ylab="Percentile",
 153    col.hor='blue', col.vert='blue', pch='.', col.points='blue', add=TRUE)
 154 plot(d_m1_may, xlim=c(0,180), ylim=c(0,1), xlab="Days to Resolve", ylab="Percentile",
 155    col.hor='green', col.vert='green', pch='.', col.points='green', add=TRUE)
 156
 157 weeks <- c(0,7,14,21,28,60,90,120,150,180)
 158 axis(1, labels=weeks, at=weeks)
 159 percentages <- c(0,0.25, 0.5, 0.75, 0.85, 0.95, 1)
 160 axis(2, las=1, labels=percentages, at=percentages)
 161
 162 abline(v=c(7,14,21,28), col='grey80', lty=2)
 163 abline(h=c(0.5, 0.6, 0.75, 0.85, 0.95 ), col='grey80', lty=2)
 164 abline(v=c(91), col='grey80', lty=2)
 165
 166
 167 legend(100, 0.1,
 168        cex=0.7,
 169        legend=c("Typical MyOps -- July2009-Feb2010", "Notice Bug -- Oct2008-Mar2009", "Kernel Bug -- May2008-Sept2008"),
 170        pch=c('-', '-', '-'),
 171        col=c('red', 'blue', 'green'),
 172        lty=c(1, 1, 1), merge=T)
 173 end_image()