Add telemetry for CRLite
Categories
(Core :: Security: PSM, task, P3)
Tracking
()
People
(Reporter: mgoodwin, Unassigned)
References
(Blocks 1 open bug)
Details
(Whiteboard: [psm-backlog])
We need to collect some telemetry on CRLite - initially this will include revocation numbers (would a cert be blocked under the new mechanism) but we will likely want mnore detail.
We need to define what we need to collect.
Updated•6 years ago
|
Updated•6 years ago
|
Comment 1•5 years ago
|
||
The first two probes are landing in Bug 1488865.
This bug should cover the remaining queries. The first ones we've identified are:
-
CRLITE_QUERY_TIME_MS :: timestamp
How long it takes to query the bloom filter and get a result. -
CRLITE_QUERY_RESULT :: { hitRevoked, hitValid, miss }
Whether results are revoked, valid, or there was no result
The next two queries are for running CRLite and OCSP simultaneously to compare performance:
-
CRLITE_OCSP_AGREEMENT :: { agree, ocspRevokedCrliteValid, ocspValidCrliteRevoked, ocspErrorCrliteValid, ocspErrorCrliteRevoked, ocspRevokedCrliteMiss, ocspValidCrliteMiss, ocspErrorCrliteMiss }
When a result returns from both the bloom filter and OCSP, do they agree or disagree, and if they disagree, how? -
CRLITE_OCSP_SPEEDUP_MS :: signed number of milliseconds CRLite is faster (or slower, if negative) (edited)
When a result returns from both the bloom filter and OCSP, did the bloom filter respond faster, and if so how much faster?
Comment 2•5 years ago
|
||
Telemetry doesn't support signed measures, so you may wish to recast CRLITE_OCSP_SPEEDUP as being either proportional to each other (Accumulate(125)
if CRLite was 125% faster than OCSP, for instance) or to adjust the "0" to be somewhat higher (Accumulate(FAKE_ZERO + (-250))
if CRLite was 250ms slower than OCSP. FAKE_ZERO could be 1000 or something).
I recommend playing with the histogram simulator to see what limitations histograms have. You can even supply your own custom data.
Comment 3•5 years ago
|
||
Note: Data-review request in https://bugzilla.mozilla.org/attachment.cgi?id=9062379
Data-review+ in https://bugzilla.mozilla.org/show_bug.cgi?id=1488865#c5
Updated•5 years ago
|
Description
•