statistics/myops_restoration.r

   1 source("functions.r");
   2
   3
   4 nsh <- read.csv('node_status_history.csv', sep=',', header=TRUE)
   5
   6 # system("./harvest_nodehistory.py > node_status_history_nopcu.csv")
   7 nsh_nopcu <- read.csv('node_status_history_nopcu.csv', sep=',', header=TRUE)
   8
   9 nsh_m1 <- read.csv('node_status_history_m1.csv', sep=',', header=TRUE)
  10 # system("stats-m1/harvest_nodehistory_m1.py > ./node_status_history_m1_nopcu.csv")
  11 nsh_m1_nopcu_total <- read.csv('node_status_history_m1_nopcu_total.csv', sep=',', header=TRUE)
  12 nsh_m1_nopcu_notice <- read.csv('node_status_history_m1_nopcu.csv', sep=',', header=TRUE)
  13 nsh_m1_nopcu_kernel <- read.csv('node_status_history_m1_nopcu_may08sep08.csv', sep=',', header=TRUE)
  14
  15 node_hist_image <- function (t, year, from, to, max=0, type="week", title="")
  16 {
  17     dates <-seq(as.Date(from), as.Date(to), type)
  18     months <- format(dates, "%b-%d")
  19     hbreaks<-unclass(as.POSIXct(dates))
  20
  21     image <- matrix(data=0, nrow=max(as.numeric(t$hostname)), ncol=length(hbreaks))
  22
  23     for ( i in seq(1, length(hbreaks)) )
  24     {
  25         # find the range : d plus a day
  26         d <- hbreaks[i]
  27         d_end <- d+60*60*24
  28         # find unique hosts in this day range
  29         t_sub <- t[which(t$start > d & t$start <= d_end & t$status == 'down'),]
  30         unique_hosts <- unique(t_sub$hostname)
  31         if (length(unique_hosts) == 0 ) { next }
  32
  33         host_n_list <- unique_hosts
  34         host_s_list <- as.character(unique_hosts)
  35
  36         for ( hi in seq(1, length(unique_hosts))  )
  37         {
  38             host_s <- host_s_list[hi]
  39             host_n <- host_n_list[hi]
  40             # events for this host after d (to avoid already identified events)
  41             ev <- t[which(t$hostname == host_s & t$start > d ),]
  42             print (sprintf("events length for host %s %s", host_s, length(ev$start)));
  43             # get down events for this host
  44             down_ev_index_list <- which(ev$status == 'down')
  45             for ( e_i in down_ev_index_list )
  46             {
  47                 if ( e_i == length(ev$status) ) {
  48                     # then the node ends down, so fill in the rest with 1.
  49                     for ( j in seq(i,length(hbreaks)) ) {
  50                         image[host_n,j] <- 1
  51                     }
  52                 } else {
  53                     # then there is a subsequent 'good' event
  54                     good_ev <- ev[e_i+1,]
  55                     down_ev <- ev[e_i,]
  56                     dbreaks <- seq(d,good_ev$start+60*60*24,60*60*24)
  57                     # for every index for time d, to good_ev$start
  58                     l<-length(dbreaks)
  59                     print (sprintf("length %s",l));
  60                     for ( j in seq(i,i+l) )
  61                     {
  62                         image[host_n,j] <- 1
  63                     }
  64                 }
  65             }
  66         }
  67     }
  68     myImagePlot(image, xLabels=months, yLabels=c(""), title=title)
  69     return (image);
  70 }
  71
  72
  73
  74 node_hist_dist <- function (t, year, from, to, max=0, type="week", title="")
  75 {
  76     dates <-seq(as.Date(from), as.Date(to), type)
  77     months <- format(dates, "%b-%d")
  78     hbreaks<-unclass(as.POSIXct(dates))
  79     current_ts <- unclass(as.POSIXct(Sys.Date()))
  80
  81     dist <- NULL
  82
  83     unique_hosts <- unique(t$hostname)
  84     host_n_list <- unique_hosts
  85     host_s_list <- as.character(unique_hosts)
  86
  87     down <- 0
  88
  89     for ( hi in seq(1, length(unique_hosts))  )
  90     {
  91         host_s <- host_s_list[hi]
  92         host_n <- host_n_list[hi]
  93         # events for this host after d (to avoid already identified events)
  94         ev <- t[which( t$hostname == host_s ),]
  95         print (sprintf("events length for host %s %s", host_s, length(ev$start)));
  96         # get down events for this host
  97         down_ev_index_list <- which(ev$status == 'down')
  98         for ( e_i in down_ev_index_list )
  99         {
 100             # when equal, there is no resolution so leave it as down
 101             if ( e_i != length(ev$status) ) {
 102                 good_ev <- ev[e_i+1,]
 103                 down_ev <- ev[e_i,]
 104                 dist <- c(dist, good_ev$start - down_ev$start)
 105             } else if ( e_i == length(ev$status) && length(ev$status) == 1) {
 106                 print (sprintf("DOWN FOREVER! %s", length(ev$start) ))
 107                 down <- down + 1
 108                 dist <- c(dist, 10*current_ts - ev$start)
 109             }
 110         }
 111     }
 112     print(down);
 113     return (dist);
 114 }
 115
 116
 117
 118 # data collected from M2 pickle files
 119 dnc <- read.csv('daily-available-node-count.csv', sep=',', header=TRUE)
 120
 121 dnc2<-add_timestamp(dnc)
 122
 123 tstamp_08 <-unclass(as.POSIXct("2008-05-07", origin="1970-01-01"))[1]
 124 dnc2 <- dnc2[which( dnc2$start >  tstamp_08 ),]
 125
 126
 127 dates <-seq(as.Date('2008-05-07'), as.Date('2009-05-07'), 'week')
 128 months <- format(dates, "%b")
 129 hbreaks<-unclass(as.POSIXct(dates))
 130
 131 x_start<-unclass(as.POSIXct("2008-05-01", origin="1970-01-01"))[1]
 132 x_end  <-unclass(as.POSIXct("2009-06-1", origin="1970-01-01"))[1]
 133
 134
 135 tstamp_0510 <-abline_at_date("2008-05-10", col='grey20', lty=0, height=570)
 136 # dates takes from reboot_image() output for API events.
 137 # green
 138 tstamp_0610 <-abline_at_date("2008-06-10", col='grey40', lty=5, height=570)
 139 tstamp_0815 <-abline_at_date("2008-08-15", col='grey70', lty=1, height=570)
 140
 141 # red
 142 #tstamp_0905 <-abline_at_date("2008-09-05", col='grey70', height=570)
 143 tstamp_0924 <-abline_at_date("2008-09-24", col='grey70', lty=1, height=570)
 144 tstamp_1015 <-abline_at_date("2008-10-15", col='grey40', lty=5, height=570)
 145 # blue
 146 #tstamp_1105 <-abline_at_date("2008-11-05", col='white', lty=2, height=570)
 147 #tstamp_1214 <-abline_at_date("2008-12-14", col='grey70', height=570)
 148 tstamp_0223 <-abline_at_date("2009-02-23", col='grey70', height=570)
 149 # red
 150 #tstamp_0313 <-abline_at_date("2009-03-13", col='grey70', height=570)
 151
 152 start_image("myops_restore_nopcu.png")
 153 par(mfrow=c(2,1))
 154 par(mai=c(.9,.8,.1,.1))
 155 plot(dnc2$start[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0815 & dnc2$start <= tstamp_1015) )],
 156     dnc2$available[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0815 & dnc2$start <= tstamp_1015) )],
 157     type='l', col='red', ylim=c(0,600), xlim=c(x_start, x_end),
 158     xlab="", ylab="Online Node Count", axes=F)
 159
 160 lines(dnc2$start[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0223) )],
 161     dnc2$available[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0223) )],
 162     type='l', col='red')
 163
 164 lines(dnc2$start[which(!is.na(dnc2$available) & dnc2$start > tstamp_1015 & dnc2$start <= tstamp_0223)], dnc2$available[which(!is.na(dnc2$available)& dnc2$start > tstamp_1015 & dnc2$start <= tstamp_0223)], lty=2, type='l', col='blue')
 165
 166 lines(dnc2$start[which(!is.na(dnc2$available) & dnc2$start > tstamp_0510 & dnc2$start <= tstamp_0815)], dnc2$available[which(!is.na(dnc2$available)& dnc2$start > tstamp_0510 & dnc2$start <= tstamp_0815)], lty=3, type='l', col='darkgreen')
 167
 168 #lines(dnc2$start[which(!is.na(dnc2$available))], dnc2$available[which(!is.na(dnc2$available))],
 169 #type='l', col='red', ylim=c(0,1000))
 170 axis(2, las=1)
 171 axis(1, cex.axis=0.7, labels=months, at=hbreaks)
 172
 173
 174
 175 tstamp_0510 <-abline_at_date("2008-05-10", col='grey20', lty=0, height=570)
 176 # dates takes from reboot_image() output for API events.
 177 # green
 178 tstamp_0610 <-abline_at_date("2008-06-10", col='grey40', lty=5, height=570)
 179 tstamp_0815 <-abline_at_date("2008-08-15", col='grey70', lty=1, height=570)
 180
 181 # red
 182 #tstamp_0905 <-abline_at_date("2008-09-05", col='grey70', height=570)
 183 tstamp_0924 <-abline_at_date("2008-09-24", col='grey70', lty=1, height=570)
 184 tstamp_1015 <-abline_at_date("2008-10-15", col='grey40', lty=5, height=570)
 185 # blue
 186 #tstamp_1105 <-abline_at_date("2008-11-05", col='white', lty=2, height=570)
 187 #tstamp_1214 <-abline_at_date("2008-12-14", col='grey70', height=570)
 188 tstamp_0223 <-abline_at_date("2009-02-23", col='grey70', height=570)
 189 # red
 190 #tstamp_0313 <-abline_at_date("2009-03-13", col='grey70', height=570)
 191
 192 #text(x=c(tstamp_0610+(tstamp_0815-tstamp_0610)/2,
 193 #         tstamp_0815+(tstamp_0905-tstamp_0815)/2,
 194 #         tstamp_0924+(tstamp_1015-tstamp_0924)/2,
 195 #         tstamp_1015+(tstamp_1214-tstamp_1015)/2,
 196 #         tstamp_1214+(tstamp_0223-tstamp_1214)/2,
 197 #         tstamp_0223+(tstamp_0313-tstamp_0223)/2),
 198 #     y=c(0),
 199 #     labels=c("bug1", 'fix1', 'fix2', 'fix3', 'bug2', 'fix4')) #, 'fix 2', 'fix 3', 'fix 4'))
 200
 201 text(x=c( tstamp_0815,
 202          tstamp_0924,
 203          tstamp_0223),
 204      y=c(610),
 205      adj=c(1, 0.5),
 206      labels=c('fix1', 'fix2', 'fix3'))
 207
 208
 209 text(x=c(tstamp_0510-(60*60*24*10),
 210         tstamp_0610,
 211         tstamp_1015),
 212      adj=c(0, 0.5),
 213      y=c(610),
 214      labels=c('Events:', 'bug1', 'bug2'))
 215
 216 mtext("2008                                 2009", 1,2)
 217 legend(unclass(as.POSIXct("2009-02-23", origin="1970-01-01"))[1], 200,
 218         cex=0.7,
 219         legend=c("Typical MyOps", "Bug1", "Bug2", 'Bug Added', 'Fix Added'),
 220         pch=c('-', '-', '-'),
 221         col=c('red', 'darkgreen', 'blue', 'grey20', 'grey70'),
 222         lty=c(1, 3, 2, 5, 1), merge=T)
 223
 224         #legend=c("Registered", "Online", 'Kernel Update', 'MyOps Event'),
 225         #pch=c('-', '-', '-', '-'),
 226         #col=c('blue', 'red', 'grey20', 'grey70'),
 227         #lty=c(1, 1, 2, 1), merge=T)
 228
 229 ###################################
 230
 231 t_0815 <- unclass(as.POSIXct("2008-08-15", origin="1970-01-01"))[1]
 232 t_0905 <- unclass(as.POSIXct("2008-09-05", origin="1970-01-01"))[1]
 233
 234 t_0924 <- unclass(as.POSIXct("2008-09-24", origin="1970-01-01"))[1]
 235 t_1015 <- unclass(as.POSIXct("2008-10-15", origin="1970-01-01"))[1]
 236
 237 t_0223 <- unclass(as.POSIXct("2009-02-23", origin="1970-01-01"))[1]
 238 t_0313 <- unclass(as.POSIXct("2009-03-13", origin="1970-01-01"))[1]
 239
 240 nsh_m1_short <- nsh_m1_nopcu_total[which(
 241         (nsh_m1_nopcu_total$start > t_0815 & nsh_m1_nopcu_total$start <= t_0313) ),]
 242 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-05-01', '2009-05-22', 0, 'day')
 243 d_m1_total<- ecdf(nsh_dist_m1/(60*60*24))
 244
 245 # NOTE: something happened betweeen 10-2 and 10-3
 246 # NOTICE BUG
 247 t_1015 <- unclass(as.POSIXct("2008-10-15", origin="1970-01-01"))[1]
 248 t_0224 <- unclass(as.POSIXct("2009-02-24", origin="1970-01-01"))[1]
 249 nsh_m1_short <- nsh_m1_nopcu_notice[which(nsh_m1_nopcu_notice$start > t_1015 & nsh_m1_nopcu_notice$start <= t_0224),]
 250 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-10-01', '2009-03-22', 0, 'day')
 251 d_m1_notice_bug <- ecdf(nsh_dist_m1/(60*60*24))
 252
 253
 254 # KERNEL BUG
 255 t_0530 <- unclass(as.POSIXct("2008-05-30", origin="1970-01-01"))[1]
 256 t_0815 <- unclass(as.POSIXct("2008-08-15", origin="1970-01-01"))[1]
 257 nsh_m1_short <- nsh_m1_nopcu_kernel[which(nsh_m1_nopcu_kernel$start > t_0530 & nsh_m1_nopcu_kernel$start <= t_0815),]
 258 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-05-10', '2008-08-15', 0, 'day')
 259 d_m1_kernel_bug <- ecdf(nsh_dist_m1/(60*60*24))
 260
 261
 262 # d<-ecdf(nsh_dist[which(nsh_dist/(60*60*24) < 90 )]/(60*60*24)),
 263 # 180 ~= 6 months.
 264 par(mai=c(.9,.9,.1,.3))
 265 #plot(d, xlim=c(0,180), ylim=c(0,1), axes=F, xlab="Days to Resolve", ylab="Percentile",
 266 #   col.hor='red', col.vert='red', pch='.', col.points='red', main="")
 267
 268 x_lim_max <- 150
 269
 270 plot(d_m1_total, xlim=c(0,x_lim_max), ylim=c(0,1), axes=F, xlab="Days to Resolve",
 271     ylab="Fraction of Offline Nodes Restored", col.hor='red', col.vert='red', pch='.',
 272     col.points='red', main="")
 273
 274 plot(d_m1_notice_bug, xlim=c(0,x_lim_max), ylim=c(0,1), xlab="Days to Resolve",
 275     col.hor='blue', col.vert='blue', pch='.',
 276     col.points='blue', lty=2, add=TRUE)
 277
 278 plot(d_m1_kernel_bug, xlim=c(0,x_lim_max), ylim=c(0,1), xlab="Days to Resolve",
 279     col.hor='darkgreen', col.vert='darkgreen', pch='.',
 280     col.points='darkgreen', lty=3, add=TRUE)
 281
 282 weeks <- c(0,7,14,21,28,60,90,120,150,180)
 283 axis(1, labels=weeks, at=weeks)
 284 percentages <- c(0,0.25, 0.5, 0.75, 0.85, 0.95, 1)
 285 axis(2, las=1, labels=percentages, at=percentages)
 286
 287 abline(v=c(7,14,21,28), col='grey80', lty=2)
 288 abline(h=c(0.5, 0.6, 0.75, 0.85, 0.95 ), col='grey80', lty=2)
 289 abline(v=c(91), col='grey80', lty=2)
 290
 291
 292 legend(92, 0.25,
 293        cex=0.7,
 294        legend=c("Typical MyOps", "Only Notices", "No Notices"),
 295        pch=c('-', '-', '-'),
 296        col=c('red', 'blue', 'darkgreen'),
 297        lty=c(1, 2, 3), merge=T)
 298
 299 end_image()