localR/process_status.R

In MazamaScience/MazamaProductionUtils: Utility Functions to Track Server Performance

# Demonstration of how to get top processes

library(dplyr)
library(stringr)
library(ggplot2)

# Gather stats -----------------------------------------------------------------

# NOTE:  ps adjusts column spacing depending on the contents. Thus, we need
# NOTE:  to ask for results with variable string length (e.g. user, command)
# NOTE:  separately to make things easier to parse.

# ----- Process IDs, CPU and memory -----
#
# $ ps -axo pid,ppid,pgid,pcpu,pmem | head -3
#   PID  PPID  PGID  %CPU %MEM
#     1     0     1   1.9  0.1
#    40     1    40   0.0  0.0
#<123456789012345678901234567890>
#<         111111111122222222223>

ps_raw <- sys::exec_internal("ps", "-axo pid,ppid,pgid,pcpu,pmem")
ps_char <- base::rawToChar(ps_raw$stdout)
col_positions <- readr::fwf_positions(
  start = c(1,7,13,19,25),
  end = c(5,11,17,23,28),
  col_names = c("pid","ppid","pgid","pcpu","pmem")
)
df1 <- readr::read_fwf(
  ps_char,
  col_positions,
  col_types = c('cccdd'),
  skip = 1
)

# ----- User information -----
#
# $ ps -axo pid,uid,user | head -3
#   PID   UID USER
#     1     0 root            
#    40     0 root            
#<123456789012345678901234567890>
#<         111111111122222222223>

ps_raw <- sys::exec_internal("ps", "-axo pid,uid,user")
ps_char <- base::rawToChar(ps_raw$stdout)
col_positions <- readr::fwf_positions(
  start = c(1,7,13),
  end = c(5,11,NA),
  col_names = c("pid","uid","user")
)
df2 <- readr::read_fwf(
  ps_char,
  col_positions,
  col_types = c('ccc'),
  skip = 1
)

# ----- Command -----
#
# $ ps -axo pid,command | head -3
#   PID COMMAND
#     1 /sbin/launchd
#    40 /usr/sbin/syslogd
#<123456789012345678901234567890>
#<         111111111122222222223>

ps_raw <- sys::exec_internal("ps", "-axo pid,command")
ps_char <- base::rawToChar(ps_raw$stdout)
col_positions <- readr::fwf_positions(
  start = c(1,7),
  end = c(5,NA),
  col_names = c("pid","command")
)
df3 <- readr::read_fwf(
  ps_char,
  col_positions,
  col_types = c('cc'),
  skip = 1
)

# ----- CPU time -----
#
# NOTE:  By itself because it has optional [dd-] days
#
# $ ps -axo pid,cputime | head -3
#   PID     TIME
#     1 00:02:38
#     2 00:00:00
#<123456789012345678901234567890>
#<         111111111122222222223>

ps_raw <- sys::exec_internal("ps", "-axo pid,cputime")
ps_char <- base::rawToChar(ps_raw$stdout)
col_positions <- readr::fwf_positions(
  start = c(1,7),
  end = c(5,NA),
  col_names = c("pid","cputime")
)
df4 <- readr::read_fwf(
  ps_char,
  col_positions,
  col_types = c('cc'),
  skip = 1
)

# Convert into seconds
# Get the day part by matching any "..-*" and grouping the first two characters
dayPart <- str_match(df4$cputime, '(..)-')[,2]
# Get the time part by stripping off any day part
hmsPart <- stringr::str_replace(df4$cputime, "^.*-", "")
days <- as.integer(dayPart)
daySecs <- 60*60*24 * replace(days, is.na(days), 0)
# Use lubridate::hms()
hmsSecs <- as.integer(lubridate::as.duration(lubridate::hms(hmsPart)))
# NOTE:  On a Mac, hmsPart may or may not have hours which will cause
# NOTE:  lubridate::hms() to return NA. Just assign a small 100 secs here.
hmsSecs <- replace(hmsSecs, is.na(hmsSecs), 100) 
df4$cpusecs <- daySecs + hmsSecs
# Remove cputime
###df4$cputime <- NULL # keep it around while debugging

# ----- Elapsed time -----
#
# NOTE:  By itself because it has optional [dd-] days
#
# $ ps -axo pid,etime | head -3
#   PID     TIME
#     1 12-00:02:38
#     2 12-00:00:00
#<123456789012345678901234567890>
#<         111111111122222222223>

ps_raw <- sys::exec_internal("ps", "-axo pid,etime")
ps_char <- base::rawToChar(ps_raw$stdout)
col_positions <- readr::fwf_positions(
  start = c(1,7),
  end = c(5,NA),
  col_names = c("pid","etime")
)
df5 <- readr::read_fwf(
  ps_char,
  col_positions,
  col_types = c('cc'),
  skip = 1
)

# Convert into seconds
# Get the day part by matching any "..-*" and grouping the first two characters
dayPart <- str_match(df5$etime, '(..)-')[,2]
# Get the time part by stripping off any day part
hmsPart <- stringr::str_replace(df5$etime, "^.*-", "")
days <- as.integer(dayPart)
daySecs <- 60*60*24 * replace(days, is.na(days), 0)
hmsSecs <- as.integer(lubridate::as.duration(lubridate::hms(hmsPart)))
# NOTE:  On a Mac, hmsPart may or may not have hours which will cause
# NOTE:  lubridate::hms() to return NA. Just assign a small 100 secs here.
hmsSecs <- replace(hmsSecs, is.na(hmsSecs), 100) 
df5$esecs <- daySecs + hmsSecs
# Remove etime
###df5$etime <- NULL # keep it around while debugging


# ----- Combine dataframes -----

ps_DF <- left_join(df1, df2)
ps_DF <- left_join(ps_DF, df3)
ps_DF <- left_join(ps_DF, df4)
ps_DF <- left_join(ps_DF, df5)


# Work with the results --------------------------------------------------------

# ----- Who is using how much CPU?

cpu_hogs <- ps_DF %>% 
  select(pcpu,user) %>% 
  group_by(user) %>% 
  summarise(cpu_usage = sum(pcpu)) %>%
  arrange(desc(cpu_usage))

head(cpu_hogs)

# ----- Which non-root users have long running processes?

long_running <- ps_DF %>% 
  select(esecs, user, command) %>% 
  filter(user != "root") %>% 
  arrange(desc(esecs))

head(long_running)