def query(self):
bs = block_size.c.block_size
query = powa_getstatdata_sample("db")
query = query.alias()
c = query.c
return (select([
extract("epoch", c.ts).label("ts"),
(sum(c.runtime) / greatest(sum(c.calls), 1)).label("avg_runtime"),
total_read(c),
total_hit(c)
]).where(c.calls != None).group_by(c.ts, bs).order_by(
c.ts).params(samples=100))
def query(self):
bs = block_size.c.block_size
query = powa_getstatdata_sample("db")
query = query.alias()
c = query.c
return (select([
extract("epoch", c.ts).label("ts"),
(sum(c.runtime) / greatest(sum(c.calls), 1)).label("avg_runtime"),
total_read(c),
total_hit(c)])
.where(c.calls != None)
.group_by(c.ts, bs)
.order_by(c.ts)
.params(samples=100))
def query(self):
# Fetch the base query for sample, and filter them on the database
bs = block_size.c.block_size
subquery = powa_getstatdata_sample("db")
# Put the where clause inside the subquery
subquery = subquery.where(column("datname") == bindparam("database"))
query = subquery.alias()
c = query.c
return (select([
to_epoch(c.ts),
(sum(c.runtime) / greatest(sum(c.calls), 1.)).label("avg_runtime"),
total_read(c),
total_hit(c)
]).where(c.calls != None).group_by(c.ts, bs).order_by(
c.ts).params(samples=100))
def query(self):
# Fetch the base query for sample, and filter them on the database
bs = block_size.c.block_size
subquery = powa_getstatdata_sample("db")
# Put the where clause inside the subquery
subquery = subquery.where(column("datname") == bindparam("database"))
query = subquery.alias()
c = query.c
return (select([
to_epoch(c.ts),
(sum(c.runtime) / greatest(sum(c.calls), 1.)).label("avg_runtime"),
total_read(c),
total_hit(c)])
.where(c.calls != None)
.group_by(c.ts, bs)
.order_by(c.ts)
.params(samples=100))
def query(self):
# Fetch the base query for sample, and filter them on the database
bs = block_size.c.block_size
subquery = powa_getstatdata_sample("db", bindparam("server"))
# Put the where clause inside the subquery
subquery = subquery.where(column("datname") == bindparam("database"))
query = subquery.alias()
c = query.c
cols = [
c.srvid,
to_epoch(c.ts),
(sum(c.calls) /
greatest(extract("epoch", c.mesure_interval), 1)).label("calls"),
(sum(c.runtime) / greatest(sum(c.calls), 1.)).label("avg_runtime"),
(sum(c.runtime) /
greatest(extract("epoch", c.mesure_interval), 1)).label("load"),
total_read(c),
total_hit(c)
]
from_clause = query
if self.has_extension(self.path_args[0], "pg_stat_kcache"):
# Add system metrics from pg_stat_kcache,
kcache_query = kcache_getstatdata_sample("db")
kc = inner_cc(kcache_query)
kcache_query = (kcache_query.where(
(kc.srvid == bindparam("server"))
& (kc.datname == bindparam("database"))).alias())
kc = kcache_query.c
def sum_per_sec(col):
ts = extract("epoch", greatest(c.mesure_interval, '1 second'))
return (sum(col) / ts).label(col.name)
total_sys_hit = (total_read(c) - sum(kc.reads) /
greatest(extract("epoch", c.mesure_interval), 1.)
).label("total_sys_hit")
total_disk_read = (sum(kc.reads) /
greatest(extract("epoch", c.mesure_interval),
1.)).label("total_disk_read")
minflts = sum_per_sec(kc.minflts)
majflts = sum_per_sec(kc.majflts)
# nswaps = sum_per_sec(kc.nswaps)
# msgsnds = sum_per_sec(kc.msgsnds)
# msgrcvs = sum_per_sec(kc.msgrcvs)
# nsignals = sum_per_sec(kc.nsignals)
nvcsws = sum_per_sec(kc.nvcsws)
nivcsws = sum_per_sec(kc.nivcsws)
cols.extend([
total_sys_hit,
total_disk_read,
minflts,
majflts,
# nswaps, msgsnds, msgrcvs, nsignals,
nvcsws,
nivcsws
])
from_clause = from_clause.join(kcache_query,
kcache_query.c.ts == c.ts)
return (select(cols).select_from(from_clause).where(
c.calls is not None).group_by(c.srvid, c.ts, bs,
c.mesure_interval).order_by(
c.ts).params(samples=100))
def query(self):
query = powa_getstatdata_sample("query")
query = query.where(
(column("datname") == bindparam("database")) &
(column("queryid") == bindparam("query")))
query = query.alias()
c = query.c
total_blocks = ((c.shared_blks_read + c.shared_blks_hit)
.label("total_blocks"))
def bps(col):
ts = extract("epoch", greatest(c.mesure_interval, '1 second'))
return (mulblock(col) / ts).label(col.name)
cols = [to_epoch(c.ts),
c.rows,
c.calls,
case([(total_blocks == 0, 0)],
else_=cast(c.shared_blks_hit, Numeric) * 100 /
total_blocks).label("hit_ratio"),
bps(c.shared_blks_read),
bps(c.shared_blks_hit),
bps(c.shared_blks_dirtied),
bps(c.shared_blks_written),
bps(c.local_blks_read),
bps(c.local_blks_hit),
bps(c.local_blks_dirtied),
bps(c.local_blks_written),
bps(c.temp_blks_read),
bps(c.temp_blks_written),
c.blk_read_time,
c.blk_write_time,
(c.runtime / greatest(c.calls, 1)).label("avg_runtime")]
from_clause = query
if self.has_extension("pg_stat_kcache"):
# Add system metrics from pg_stat_kcache,
# and detailed hit ratio.
kcache_query = kcache_getstatdata_sample()
kc = inner_cc(kcache_query)
kcache_query = (
kcache_query
.where(kc.queryid == bindparam("query"))
.alias())
kc = kcache_query.c
sys_hits = (greatest(mulblock(c.shared_blks_read) -
kc.reads, 0)
.label("kcache_hitblocks"))
sys_hitratio = (cast(sys_hits, Numeric) * 100 /
mulblock(total_blocks))
disk_hit_ratio = (kc.reads /
mulblock(total_blocks))
total_time = greatest(c.runtime, 1);
# Rusage can return values > real time due to sampling bias
# aligned to kernel ticks. As such, we have to clamp values to 100%
total_time_percent = lambda x: least(100, (x * 100) /
total_time)
cols.extend([
kc.reads,
kc.writes,
total_time_percent(kc.user_time * 1000).label("user_time"),
total_time_percent(kc.system_time * 1000).label("system_time"),
greatest(total_time_percent(
c.runtime - (kc.user_time + kc.system_time) *
1000), 0).label("other_time"),
case([(total_blocks == 0, 0)],
else_=disk_hit_ratio).label("disk_hit_ratio"),
case([(total_blocks == 0, 0)],
else_=sys_hitratio).label("sys_hit_ratio")])
from_clause = from_clause.join(
kcache_query,
kcache_query.c.ts == c.ts)
else:
cols.extend([
case([(total_blocks == 0, 0)],
else_=cast(c.shared_blks_read, Numeric) * 100 /
total_blocks).label("miss_ratio")
])
return (select(cols)
.select_from(from_clause)
.where(c.calls != None)
.order_by(c.ts)
.params(samples=100))
def query(self):
query = powa_getstatdata_sample("query", bindparam("server"))
query = query.where((column("datname") == bindparam("database"))
& (column("queryid") == bindparam("query")))
query = query.alias()
c = query.c
total_blocks = ((sum(c.shared_blks_read) +
sum(c.shared_blks_hit)).label("total_blocks"))
def get_ts():
return extract("epoch", greatest(c.mesure_interval, '1 second'))
def sumps(col):
return (sum(col) / get_ts()).label(col.name)
def bps(col):
return (mulblock(sum(col)) / get_ts()).label(col.name)
cols = [
to_epoch(c.ts),
sumps(c.rows),
sumps(c.calls),
case([(total_blocks == 0, 0)],
else_=cast(sum(c.shared_blks_hit), Numeric) * 100 /
total_blocks).label("hit_ratio"),
bps(c.shared_blks_read),
bps(c.shared_blks_hit),
bps(c.shared_blks_dirtied),
bps(c.shared_blks_written),
bps(c.local_blks_read),
bps(c.local_blks_hit),
bps(c.local_blks_dirtied),
bps(c.local_blks_written),
bps(c.temp_blks_read),
bps(c.temp_blks_written),
sumps(c.blk_read_time),
sumps(c.blk_write_time),
(sum(c.runtime) / greatest(sum(c.calls), 1)).label("avg_runtime")
]
from_clause = query
if self.has_extension(self.path_args[0], "pg_stat_kcache"):
# Add system metrics from pg_stat_kcache,
# and detailed hit ratio.
kcache_query = kcache_getstatdata_sample("query")
kc = inner_cc(kcache_query)
kcache_query = (kcache_query.where(
(kc.srvid == bindparam("server"))
& (kc.datname == bindparam("database"))
& (kc.queryid == bindparam("query"))).alias())
kc = kcache_query.c
sys_hits = (greatest(
mulblock(sum(c.shared_blks_read)) - sum(kc.reads),
0).label("kcache_hitblocks"))
sys_hitratio = (cast(sys_hits, Numeric) * 100 /
mulblock(total_blocks))
disk_hit_ratio = (sum(kc.reads) * 100 / mulblock(total_blocks))
total_time = greatest(sum(c.runtime), 1)
# Rusage can return values > real time due to sampling bias
# aligned to kernel ticks. As such, we have to clamp values to 100%
total_time_percent = lambda x: least(100, (x * 100) / total_time)
def per_sec(col):
ts = extract("epoch", greatest(c.mesure_interval, '1 second'))
return (sum(col) / ts).label(col.name)
cols.extend([
per_sec(kc.reads),
per_sec(kc.writes),
per_sec(kc.minflts),
per_sec(kc.majflts),
# per_sec(kc.nswaps),
# per_sec(kc.msgsnds),
# per_sec(kc.msgrcvs),
# per_sec(kc.nsignals),
per_sec(kc.nvcsws),
per_sec(kc.nivcsws),
total_time_percent(sum(kc.user_time) *
1000).label("user_time"),
total_time_percent(sum(kc.system_time) *
1000).label("system_time"),
greatest(
total_time_percent(
sum(c.runtime) -
((sum(kc.user_time) + sum(kc.system_time)) * 1000)),
0).label("other_time"),
case([(total_blocks == 0, 0)],
else_=disk_hit_ratio).label("disk_hit_ratio"),
case([(total_blocks == 0, 0)],
else_=sys_hitratio).label("sys_hit_ratio")
])
from_clause = from_clause.join(
kcache_query,
and_(kcache_query.c.ts == c.ts,
kcache_query.c.queryid == c.queryid,
kcache_query.c.userid == c.userid,
kcache_query.c.dbid == c.dbid))
else:
cols.extend([
case([(total_blocks == 0, 0)],
else_=cast(sum(c.shared_blks_read), Numeric) * 100 /
total_blocks).label("miss_ratio")
])
return (select(cols).select_from(from_clause).where(
c.calls != '0').group_by(c.ts, block_size.c.block_size,
c.mesure_interval).order_by(
c.ts).params(samples=100))
def query(self):
bs = block_size.c.block_size
query = powa_getstatdata_sample("db", bindparam("server"))
query = query.alias()
c = query.c
cols = [c.srvid,
extract("epoch", c.ts).label("ts"),
(sum(c.calls) / greatest(extract("epoch", c.mesure_interval),
1)).label("calls"),
(sum(c.runtime) / greatest(sum(c.calls),
1)).label("avg_runtime"),
(sum(c.runtime) / greatest(extract("epoch", c.mesure_interval),
1)).label("load"),
total_read(c),
total_hit(c)
]
if self.has_extension_version(self.path_args[0],
'pg_stat_statements', '1.8'):
cols.extend([
(sum(c.plantime) / greatest(extract("epoch", c.mesure_interval),
1)).label("planload"),
(sum(c.wal_records) / greatest(extract("epoch",
c.mesure_interval),
1)).label("wal_records"),
(sum(c.wal_fpi) / greatest(extract("epoch",
c.mesure_interval),
1)).label("wal_fpi"),
(sum(c.wal_bytes) / greatest(extract("epoch",
c.mesure_interval),
1)).label("wal_bytes")
])
from_clause = query
if self.has_extension(self.path_args[0], "pg_stat_kcache"):
# Add system metrics from pg_stat_kcache,
kcache_query = kcache_getstatdata_sample("db")
kc = inner_cc(kcache_query)
kcache_query = (
kcache_query
.where(
(kc.srvid == bindparam("server"))
)
.alias())
kc = kcache_query.c
def sum_per_sec(col):
ts = extract("epoch", greatest(c.mesure_interval, '1 second'))
return (sum(col) / ts).label(col.name)
total_sys_hit = (total_read(c) - sum(kc.reads) /
greatest(extract("epoch", c.mesure_interval), 1.)
).label("total_sys_hit")
total_disk_read = (sum(kc.reads) /
greatest(extract("epoch", c.mesure_interval), 1.)
).label("total_disk_read")
minflts = sum_per_sec(kc.minflts)
majflts = sum_per_sec(kc.majflts)
# nswaps = sum_per_sec(kc.nswaps)
# msgsnds = sum_per_sec(kc.msgsnds)
# msgrcvs = sum_per_sec(kc.msgrcvs)
# nsignals = sum_per_sec(kc.nsignals)
nvcsws = sum_per_sec(kc.nvcsws)
nivcsws = sum_per_sec(kc.nivcsws)
cols.extend([total_sys_hit, total_disk_read, minflts, majflts,
# nswaps, msgsnds, msgrcvs, nsignals,
nvcsws, nivcsws])
from_clause = from_clause.join(
kcache_query,
and_(kcache_query.c.dbid == c.dbid,
kcache_query.c.ts == c.ts))
return (select(cols)
.select_from(from_clause)
.where(c.calls != '0')
.group_by(c.srvid, c.ts, bs, c.mesure_interval)
.order_by(c.ts)
.params(samples=100))