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_total <- read.csv('node_status_history_m1_nopcu_total.csv', sep=',', header=TRUE)
  11 nsh_m1_nopcu_notice <- read.csv('node_status_history_m1_nopcu.csv', sep=',', header=TRUE)
  12 nsh_m1_nopcu_kernel <- read.csv('node_status_history_m1_nopcu_may08sep08.csv', sep=',', header=TRUE)
  13
  14 node_hist_image <- function (t, year, from, to, max=0, type="week", title="")
  15 {
  16     dates <-seq(as.Date(from), as.Date(to), type)
  17     months <- format(dates, "%b-%d")
  18     hbreaks<-unclass(as.POSIXct(dates))
  19
  20     image <- matrix(data=0, nrow=max(as.numeric(t$hostname)), ncol=length(hbreaks))
  21
  22     for ( i in seq(1, length(hbreaks)) )
  23     {
  24         # find the range : d plus a day
  25         d <- hbreaks[i]
  26         d_end <- d+60*60*24
  27         # find unique hosts in this day range
  28         t_sub <- t[which(t$start > d & t$start <= d_end & t$status == 'down'),]
  29         unique_hosts <- unique(t_sub$hostname)
  30         if (length(unique_hosts) == 0 ) { next }
  31
  32         host_n_list <- unique_hosts
  33         host_s_list <- as.character(unique_hosts)
  34
  35         for ( hi in seq(1, length(unique_hosts))  )
  36         {
  37             host_s <- host_s_list[hi]
  38             host_n <- host_n_list[hi]
  39             # events for this host after d (to avoid already identified events)
  40             ev <- t[which(t$hostname == host_s & t$start > d ),]
  41             print (sprintf("events length for host %s %s", host_s, length(ev$start)));
  42             # get down events for this host
  43             down_ev_index_list <- which(ev$status == 'down')
  44             for ( e_i in down_ev_index_list )
  45             {
  46                 if ( e_i == length(ev$status) ) {
  47                     # then the node ends down, so fill in the rest with 1.
  48                     for ( j in seq(i,length(hbreaks)) ) {
  49                         image[host_n,j] <- 1
  50                     }
  51                 } else {
  52                     # then there is a subsequent 'good' event
  53                     good_ev <- ev[e_i+1,]
  54                     down_ev <- ev[e_i,]
  55                     dbreaks <- seq(d,good_ev$start+60*60*24,60*60*24)
  56                     # for every index for time d, to good_ev$start
  57                     l<-length(dbreaks)
  58                     print (sprintf("length %s",l));
  59                     for ( j in seq(i,i+l) )
  60                     {
  61                         image[host_n,j] <- 1
  62                     }
  63                 }
  64             }
  65         }
  66     }
  67     myImagePlot(image, xLabels=months, yLabels=c(""), title=title)
  68     return (image);
  69 }
  70
  71
  72
  73 node_hist_dist <- function (t, year, from, to, max=0, type="week", title="")
  74 {
  75     dates <-seq(as.Date(from), as.Date(to), type)
  76     months <- format(dates, "%b-%d")
  77     hbreaks<-unclass(as.POSIXct(dates))
  78     current_ts <- unclass(as.POSIXct(Sys.Date()))
  79
  80     dist <- NULL
  81
  82     unique_hosts <- unique(t$hostname)
  83     host_n_list <- unique_hosts
  84     host_s_list <- as.character(unique_hosts)
  85
  86     down <- 0
  87
  88     for ( hi in seq(1, length(unique_hosts))  )
  89     {
  90         host_s <- host_s_list[hi]
  91         host_n <- host_n_list[hi]
  92         # events for this host after d (to avoid already identified events)
  93         ev <- t[which( t$hostname == host_s ),]
  94         print (sprintf("events length for host %s %s", host_s, length(ev$start)));
  95         # get down events for this host
  96         down_ev_index_list <- which(ev$status == 'down')
  97         for ( e_i in down_ev_index_list )
  98         {
  99             # when equal, there is no resolution so leave it as down
 100             if ( e_i != length(ev$status) ) {
 101                 good_ev <- ev[e_i+1,]
 102                 down_ev <- ev[e_i,]
 103                 dist <- c(dist, good_ev$start - down_ev$start)
 104             } else if ( e_i == length(ev$status) && length(ev$status) == 1) {
 105                 print (sprintf("DOWN FOREVER! %s", length(ev$start) ))
 106                 down <- down + 1
 107                 dist <- c(dist, 10*current_ts - ev$start)
 108             }
 109         }
 110     }
 111     print(down);
 112     return (dist);
 113 }
 114
 115 # Image Stuff
 116 #nsh_image <- node_hist_image(nsh, '2009', '2009-06-01', '2010-02-28', 0, 'day')
 117 #nsh_image_m1 <- node_hist_image(nsh_m1, '2009', '2008-10-01', '2009-03-28', 0, 'day')
 118
 119 #nsh_short <- nsh[which(nsh$start > unclass(as.POSIXct("2009-06-01", origin="1970-01-01"))[1]),]
 120 #nsh_short <- nsh_short[which(nsh_short$start < unclass(as.POSIXct("2009-10-31", origin="1970-01-01"))[1]),]
 121
 122 #
 123 #nsh_short <- nsh_nopcu
 124 #nsh_dist <- node_hist_dist(nsh_short, '2009', '2009-06-01', '2010-02-28', 0, 'day')
 125 #d<- ecdf(nsh_dist/(60*60*24))
 126
 127 #nsh_m1_short <- nsh_m1[which(nsh_m1$start > unclass(as.POSIXct("2008-10-01", origin="1970-01-01"))[1]),]
 128
 129 # M1 --
 130 # TOTAL
 131 t_0815 <- unclass(as.POSIXct("2008-08-15", origin="1970-01-01"))[1]
 132 t_0905 <- unclass(as.POSIXct("2008-09-05", origin="1970-01-01"))[1]
 133
 134 t_0924 <- unclass(as.POSIXct("2008-09-24", origin="1970-01-01"))[1]
 135 t_1015 <- unclass(as.POSIXct("2008-10-15", origin="1970-01-01"))[1]
 136
 137 t_0223 <- unclass(as.POSIXct("2009-02-23", origin="1970-01-01"))[1]
 138 t_0313 <- unclass(as.POSIXct("2009-03-13", origin="1970-01-01"))[1]
 139
 140 #nsh_m1_short <- nsh_m1_nopcu_total[which(
 141 #        (nsh_m1_nopcu_total$start > t_0815 & nsh_m1_nopcu_total$start <= t_0905) |
 142 #        (nsh_m1_nopcu_total$start > t_0924 & nsh_m1_nopcu_total$start <= t_1015) |
 143 #        (nsh_m1_nopcu_total$start > t_0223 & nsh_m1_nopcu_total$start <= t_0313)  ),]
 144 #
 145 #nsh_m1_short <- nsh_m1_nopcu_total[which(
 146 #        (nsh_m1_nopcu_total$start > t_0815 & nsh_m1_nopcu_total$start <= t_0905) |
 147 #        (nsh_m1_nopcu_total$start > t_0924 & nsh_m1_nopcu_total$start <= t_1015) ),]
 148
 149 nsh_m1_short <- nsh_m1_nopcu_total[which(
 150         (nsh_m1_nopcu_total$start > t_0815 & nsh_m1_nopcu_total$start <= t_0313) ),]
 151 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-05-01', '2009-05-22', 0, 'day')
 152 d_m1_total<- ecdf(nsh_dist_m1/(60*60*24))
 153
 154
 155 # NOTE: something happened betweeen 10-2 and 10-3
 156 # NOTICE BUG
 157 t_1015 <- unclass(as.POSIXct("2008-10-15", origin="1970-01-01"))[1]
 158 t_0224 <- unclass(as.POSIXct("2009-02-24", origin="1970-01-01"))[1]
 159 nsh_m1_short <- nsh_m1_nopcu_notice[which(nsh_m1_nopcu_notice$start > t_1015 & nsh_m1_nopcu_notice$start <= t_0224),]
 160 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-10-01', '2009-03-22', 0, 'day')
 161 d_m1_notice_bug <- ecdf(nsh_dist_m1/(60*60*24))
 162
 163
 164 # KERNEL BUG
 165 t_0530 <- unclass(as.POSIXct("2008-05-30", origin="1970-01-01"))[1]
 166 t_0815 <- unclass(as.POSIXct("2008-08-15", origin="1970-01-01"))[1]
 167 nsh_m1_short <- nsh_m1_nopcu_kernel[which(nsh_m1_nopcu_kernel$start > t_0530 & nsh_m1_nopcu_kernel$start <= t_0815),]
 168 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-05-10', '2008-08-15', 0, 'day')
 169 d_m1_kernel_bug <- ecdf(nsh_dist_m1/(60*60*24))
 170
 171
 172 # d<-ecdf(nsh_dist[which(nsh_dist/(60*60*24) < 90 )]/(60*60*24)),
 173 # 180 ~= 6 months.
 174 par(mfrow=c(1,1))
 175 par(mai=c(.9,.9,.1,.1))
 176 #start_image("node_history_ttr_nopcu.png")
 177 #plot(d, xlim=c(0,180), ylim=c(0,1), axes=F, xlab="Days to Resolve", ylab="Percentile",
 178 #   col.hor='red', col.vert='red', pch='.', col.points='red', main="")
 179
 180 plot(d_m1_total, xlim=c(0,180), ylim=c(0,1), axes=F, xlab="Days to Resolve",
 181     ylab="Percentile", col.hor='red', col.vert='red', pch='.',
 182     col.points='red', main="")
 183
 184 plot(d_m1_notice_bug, xlim=c(0,180), ylim=c(0,1), xlab="Days to Resolve",
 185     ylab="Percentile", col.hor='blue', col.vert='blue', pch='.',
 186     col.points='blue', add=TRUE)
 187
 188 plot(d_m1_kernel_bug, xlim=c(0,180), ylim=c(0,1), xlab="Days to Resolve",
 189     ylab="Percentile", col.hor='green', col.vert='green', pch='.',
 190     col.points='green', add=TRUE)
 191
 192 weeks <- c(0,7,14,21,28,60,90,120,150,180)
 193 axis(1, labels=weeks, at=weeks)
 194 percentages <- c(0,0.25, 0.5, 0.75, 0.85, 0.95, 1)
 195 axis(2, las=1, labels=percentages, at=percentages)
 196
 197 abline(v=c(7,14,21,28), col='grey80', lty=2)
 198 abline(h=c(0.5, 0.6, 0.75, 0.85, 0.95 ), col='grey80', lty=2)
 199 abline(v=c(91), col='grey80', lty=2)
 200
 201
 202 legend(100, 0.1,
 203        cex=0.7,
 204        legend=c("Typical MyOps -- fix1 through fix4", "Notice Bug", "Kernel Bug"),
 205        pch=c('-', '-', '-'),
 206        col=c('red', 'blue', 'green'),
 207        lty=c(1, 1, 1), merge=T)
 208
 209 end_image()