4 nsh <- read.csv('node_status_history.csv', sep=',', header=TRUE)
6 # system("./harvest_nodehistory.py > node_status_history_nopcu.csv")
7 nsh_nopcu <- read.csv('node_status_history_nopcu.csv', sep=',', header=TRUE)
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)
15 node_hist_image <- function (t, year, from, to, max=0, type="week", title="")
17 dates <-seq(as.Date(from), as.Date(to), type)
18 months <- format(dates, "%b-%d")
19 hbreaks<-unclass(as.POSIXct(dates))
21 image <- matrix(data=0, nrow=max(as.numeric(t$hostname)), ncol=length(hbreaks))
23 for ( i in seq(1, length(hbreaks)) )
25 # find the range : d plus a day
28 # find unique hosts in this day range
29 t_sub <- t[which(t$start > d & t$start <= d_end & t$status == 'down'),]
30 unique_hosts <- unique(t_sub$hostname)
31 if (length(unique_hosts) == 0 ) { next }
33 host_n_list <- unique_hosts
34 host_s_list <- as.character(unique_hosts)
36 for ( hi in seq(1, length(unique_hosts)) )
38 host_s <- host_s_list[hi]
39 host_n <- host_n_list[hi]
40 # events for this host after d (to avoid already identified events)
41 ev <- t[which(t$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$status == 'down')
45 for ( e_i in down_ev_index_list )
47 if ( e_i == length(ev$status) ) {
48 # then the node ends down, so fill in the rest with 1.
49 for ( j in seq(i,length(hbreaks)) ) {
53 # then there is a subsequent 'good' event
56 dbreaks <- seq(d,good_ev$start+60*60*24,60*60*24)
57 # for every index for time d, to good_ev$start
59 print (sprintf("length %s",l));
60 for ( j in seq(i,i+l) )
68 myImagePlot(image, xLabels=months, yLabels=c(""), title=title)
74 node_hist_dist <- function (t, year, from, to, max=0, type="week", title="")
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()))
83 unique_hosts <- unique(t$hostname)
84 host_n_list <- unique_hosts
85 host_s_list <- as.character(unique_hosts)
89 for ( hi in seq(1, length(unique_hosts)) )
91 host_s <- host_s_list[hi]
92 host_n <- host_n_list[hi]
93 # events for this host after d (to avoid already identified events)
94 ev <- t[which( t$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 )
100 # when equal, there is no resolution so leave it as down
101 if ( e_i != length(ev$status) ) {
102 good_ev <- ev[e_i+1,]
104 dist <- c(dist, good_ev$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) ))
108 dist <- c(dist, 10*current_ts - ev$start)
118 # data collected from M2 pickle files
119 dnc <- read.csv('daily-available-node-count.csv', sep=',', header=TRUE)
121 dnc2<-add_timestamp(dnc)
123 tstamp_08 <-unclass(as.POSIXct("2008-05-07", origin="1970-01-01"))[1]
124 dnc2 <- dnc2[which( dnc2$start > tstamp_08 ),]
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))
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]
134 start_image("myops_restore_nopcu.png")
136 par(mai=c(.9,.8,.1,.1))
137 plot(dnc2$start[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0815 & dnc2$start <= tstamp_1015) )],
138 dnc2$available[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0815 & dnc2$start <= tstamp_1015) )],
139 type='l', col='red', ylim=c(0,600), xlim=c(x_start, x_end),
140 xlab="", ylab="Online Node Count", axes=F)
142 lines(dnc2$start[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0223) )],
143 dnc2$available[which(!is.na(dnc2$available) & (dnc2$start > tstamp_0223) )],
146 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')
148 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')
150 #lines(dnc2$start[which(!is.na(dnc2$available))], dnc2$available[which(!is.na(dnc2$available))],
151 #type='l', col='red', ylim=c(0,1000))
153 axis(1, cex.axis=0.7, labels=months, at=hbreaks)
157 #tstamp_0510 <-abline_at_date("2008-05-10", col='grey20', lty=0, height=570)
158 # dates takes from reboot_image() output for API events.
160 tstamp_0610 <-abline_at_date("2008-06-10", col='grey40', lty=5, height=570)
161 tstamp_0815 <-abline_at_date("2008-08-15", col='grey70', lty=1, height=570)
164 #tstamp_0905 <-abline_at_date("2008-09-05", col='grey70', height=570)
165 tstamp_0924 <-abline_at_date("2008-09-24", col='grey70', lty=1, height=570)
166 tstamp_1015 <-abline_at_date("2008-10-15", col='grey40', lty=5, height=570)
168 #tstamp_1105 <-abline_at_date("2008-11-05", col='white', lty=2, height=570)
169 #tstamp_1214 <-abline_at_date("2008-12-14", col='grey70', height=570)
170 tstamp_0223 <-abline_at_date("2009-02-23", col='grey70', height=570)
172 #tstamp_0313 <-abline_at_date("2009-03-13", col='grey70', height=570)
174 #text(x=c(tstamp_0610+(tstamp_0815-tstamp_0610)/2,
175 # tstamp_0815+(tstamp_0905-tstamp_0815)/2,
176 # tstamp_0924+(tstamp_1015-tstamp_0924)/2,
177 # tstamp_1015+(tstamp_1214-tstamp_1015)/2,
178 # tstamp_1214+(tstamp_0223-tstamp_1214)/2,
179 # tstamp_0223+(tstamp_0313-tstamp_0223)/2),
181 # labels=c("bug1", 'fix1', 'fix2', 'fix3', 'bug2', 'fix4')) #, 'fix 2', 'fix 3', 'fix 4'))
183 text(x=c( tstamp_0815,
188 labels=c('fix1', 'fix2', 'fix3'))
191 text(x=c(tstamp_0510-(60*60*24*10),
196 labels=c('Events:', 'bug1', 'bug2'))
198 mtext("2008 2009", 1,2)
199 legend(unclass(as.POSIXct("2009-02-23", origin="1970-01-01"))[1], 200,
201 legend=c("Typical MyOps", "Notice Bug", "Kernel Bug", 'Bug Added', 'Fix Added'),
202 pch=c('-', '-', '-'),
203 col=c('red', 'blue', 'darkgreen', 'grey20', 'grey70'),
204 lty=c(1, 2, 3, 5, 1), merge=T)
206 #legend=c("Registered", "Online", 'Kernel Update', 'MyOps Event'),
207 #pch=c('-', '-', '-', '-'),
208 #col=c('blue', 'red', 'grey20', 'grey70'),
209 #lty=c(1, 1, 2, 1), merge=T)
211 ###################################
213 t_0815 <- unclass(as.POSIXct("2008-08-15", origin="1970-01-01"))[1]
214 t_0905 <- unclass(as.POSIXct("2008-09-05", origin="1970-01-01"))[1]
216 t_0924 <- unclass(as.POSIXct("2008-09-24", origin="1970-01-01"))[1]
217 t_1015 <- unclass(as.POSIXct("2008-10-15", origin="1970-01-01"))[1]
219 t_0223 <- unclass(as.POSIXct("2009-02-23", origin="1970-01-01"))[1]
220 t_0313 <- unclass(as.POSIXct("2009-03-13", origin="1970-01-01"))[1]
222 nsh_m1_short <- nsh_m1_nopcu_total[which(
223 (nsh_m1_nopcu_total$start > t_0815 & nsh_m1_nopcu_total$start <= t_0313) ),]
224 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-05-01', '2009-05-22', 0, 'day')
225 d_m1_total<- ecdf(nsh_dist_m1/(60*60*24))
227 # NOTE: something happened betweeen 10-2 and 10-3
229 t_1015 <- unclass(as.POSIXct("2008-10-15", origin="1970-01-01"))[1]
230 t_0224 <- unclass(as.POSIXct("2009-02-24", origin="1970-01-01"))[1]
231 nsh_m1_short <- nsh_m1_nopcu_notice[which(nsh_m1_nopcu_notice$start > t_1015 & nsh_m1_nopcu_notice$start <= t_0224),]
232 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-10-01', '2009-03-22', 0, 'day')
233 d_m1_notice_bug <- ecdf(nsh_dist_m1/(60*60*24))
237 t_0530 <- unclass(as.POSIXct("2008-05-30", origin="1970-01-01"))[1]
238 t_0815 <- unclass(as.POSIXct("2008-08-15", origin="1970-01-01"))[1]
239 nsh_m1_short <- nsh_m1_nopcu_kernel[which(nsh_m1_nopcu_kernel$start > t_0530 & nsh_m1_nopcu_kernel$start <= t_0815),]
240 nsh_dist_m1 <- node_hist_dist(nsh_m1_short, '2008', '2008-05-10', '2008-08-15', 0, 'day')
241 d_m1_kernel_bug <- ecdf(nsh_dist_m1/(60*60*24))
244 # d<-ecdf(nsh_dist[which(nsh_dist/(60*60*24) < 90 )]/(60*60*24)),
246 par(mai=c(.9,.9,.1,.3))
247 #plot(d, xlim=c(0,180), ylim=c(0,1), axes=F, xlab="Days to Resolve", ylab="Percentile",
248 # col.hor='red', col.vert='red', pch='.', col.points='red', main="")
252 plot(d_m1_total, xlim=c(0,x_lim_max), ylim=c(0,1), axes=F, xlab="Days to Resolve",
253 ylab="Fraction of Offline Nodes Restored", col.hor='red', col.vert='red', pch='.',
254 col.points='red', main="")
256 plot(d_m1_notice_bug, xlim=c(0,x_lim_max), ylim=c(0,1), xlab="Days to Resolve",
257 col.hor='blue', col.vert='blue', pch='.',
258 col.points='blue', lty=2, add=TRUE)
260 plot(d_m1_kernel_bug, xlim=c(0,x_lim_max), ylim=c(0,1), xlab="Days to Resolve",
261 col.hor='darkgreen', col.vert='darkgreen', pch='.',
262 col.points='darkgreen', lty=3, add=TRUE)
264 weeks <- c(0,7,14,21,28,60,90,120,150,180)
265 axis(1, labels=weeks, at=weeks)
266 percentages <- c(0,0.25, 0.5, 0.75, 0.85, 0.95, 1)
267 axis(2, las=1, labels=percentages, at=percentages)
269 abline(v=c(7,14,21,28), col='grey80', lty=2)
270 abline(h=c(0.5, 0.6, 0.75, 0.85, 0.95 ), col='grey80', lty=2)
271 abline(v=c(91), col='grey80', lty=2)
276 legend=c("Typical MyOps", "Notice Bug", "Kernel Bug"),
277 pch=c('-', '-', '-'),
278 col=c('red', 'blue', 'darkgreen'),
279 lty=c(1, 2, 3), merge=T)