● Based in Lyon, France
● Former designer and educator
● Loves cooking and baking
● Visual art, storytelling, photography
● @thomasdiluccio.bsky.social
Thomas di Luccio (He/Him)
DevRel @ Platform.sh,
Upsun, Blackfire
Who am I?
Slide 3
Slide 3 text
● Deterministic vs Probabilistic observability
● Collecting information and making sense out of it
● Maximizing information while minimizing overhead
A talk about:
Slide 4
Slide 4 text
Observability?
Slide 5
Slide 5 text
Observability
Ability to witness the real behavior of an
app, detect bottlenecks, improve and
control performance
Slide 6
Slide 6 text
Observability
Bionic glasses for your applications
Deterministic
profiling
● Multi-dimensional
● Observe ALL functions of the
designated requests/scripts
● Two samples can be compared
● A sample always contains reliable
pieces of information
● Metrics can be derived from samples
● Strong overhead
Slide 12
Slide 12 text
Going beyond profiling
Slide 13
Slide 13 text
Performance tests
Slide 14
Slide 14 text
No content
Slide 15
Slide 15 text
No content
Slide 16
Slide 16 text
Profiles automatically evaluate
all matching assertions
Slide 17
Slide 17 text
Custom metrics
Slide 18
Slide 18 text
No content
Slide 19
Slide 19 text
Synthetic monitoring
Slide 20
Slide 20 text
No content
Slide 21
Slide 21 text
No content
Slide 22
Slide 22 text
CI/CD pipelines
Slide 23
Slide 23 text
Assessing the consequences
of upcoming changes
Slide 24
Slide 24 text
Going even further
Slide 25
Slide 25 text
Multiple
levels
of data
collection
Profile:
Extended trace:
Monitoring trace:
Slide 26
Slide 26 text
Blackfire Monitoring
frequent collection of minimal data
Slide 27
Slide 27 text
Monitoring
sampling
rate
● Percentage of HTTP and CLI traffic to
be monitored
● Monitoring is both deterministic and
probabilistic
Slide 28
Slide 28 text
Extended
sampling
rate
● Percentage of the monitored traces
for which more information are
collected
Proactively identify existing
bottlenecks and the consequences
of upcoming changes
Slide 41
Slide 41 text
Why would need an extra tool?
Slide 42
Slide 42 text
improve the
information vs overhead
ratio
Slide 43
Slide 43 text
Correlation
between data
collection and
overhead
Profile:
data++ / overhead++
Monitoring trace:
data - - / overhead - -
Slide 44
Slide 44 text
Could we have the
best of both worlds?
Slide 45
Slide 45 text
Intermittent data
collection
Slide 46
Slide 46 text
Intermittent
data
collection continuous profiling:
data++ / overhead- -
Slide 47
Slide 47 text
Introducing
Continuous profiling
Slide 48
Slide 48 text
No content
Slide 49
Slide 49 text
No content
Slide 50
Slide 50 text
No content
Slide 51
Slide 51 text
No content
Slide 52
Slide 52 text
No content
Slide 53
Slide 53 text
No content
Slide 54
Slide 54 text
Demo
Slide 55
Slide 55 text
Continuous
profiling ● Make periodic sampling in ALL of the
requests/scripts
● Holistic observability
● Two samples cannot be compared
● A single sample contains partial
information
● Probabilistic observability
● Accurate information is derived from
a large number of samples
● Minimal overhead
Slide 56
Slide 56 text
Continuous Profiling
Holistic quality observability
with minimal overhead
Slide 57
Slide 57 text
deterministic versus
continuous profiling
Slide 58
Slide 58 text
Continuous profiling is best for:
● holistic view of application
● minimal overhead
Slide 59
Slide 59 text
Deterministic profiling is best for:
● drilling down issues
● automated performance tests
Slide 60
Slide 60 text
Deterministic
observability
Slide 61
Slide 61 text
Continuous profiling
(more to be added soon …)
Slide 62
Slide 62 text
Build you own
observability strategy
Stay in control of your
applications at all times
Slide 63
Slide 63 text
Thank you! Thomas di Luccio
DevRel Engineer, Platform.sh
[email protected]