Recently, I discovered that when an error occurs on my machine and the Feedback dialog suggests sending the report to the Tribler developer, clicking the "SEND REPORT" button does nothing. The confirmation box with the message "Successfully sent the report! Thanks for your contribution" shows, but no errors appear on the Sentry server.
Specifying the debug=True option when calling sentry_sdk.init(...) allows me to see Sentry logs configured weirdly:
The log level for the sentry log was DEBUG, which is expected for the debug=True option but too verbose. Without the debug=True option, even error-level log records from the Sentry are silenced.
All sentry logs (debug and info level including) were additionally directed to stderr, duplicating the same log messages in stdout and stderr. The formatting of these log records in the stderr output was inconsistent with Tribler logs.
With enabled Sentry logs, the following log record appeared in stderr:
[sentry] ERROR: Unexpected status code: 400 (body: b'{"detail":"invalid JSON data","causes":["invalid value: string \\"<>d27beecffcdb4caa9f914986dcc56828<>\\", expected an event identifier at line 1 column 8048"]}')
It allowed me to investigate the reason for the error. As it turns out, SentryScrubber produced a corrupted error report that was then rejected by the Sentry server.
SentryScrubber tries to remove sensitive information from the error report. It generates a set of sensitive strings with corresponding pseudo-random substitutes. For example, a username "John" may be replaced with the string "<foo>", and a machine name "JOHN-PC" may be replaced with the string "<bar>". Then, using regular expressions, SentryScrubber replaces corresponding substrings in all places of the error report.
As it turns out, the reason for the problem was that sometimes SentryReported decides to replace an empty string "" with the substitution "<>" and applies this replacement to all possible substrings. As a result, even elements added to the report by Sentry itself were transformed; for example, the identifier value
"d27beecffcdb4caa9f914986dcc56828"
was transformed into
"<>d27beecffcdb4caa9f914986dcc56828<>"
and all other string elements of the report were changed in the same way.
As a simple fix, SentryScrubber should stop considering empty strings as a potential target for replacement. After such a fix, with the current logic, SentryScrubber still has a chance to corrupt the report if some sensitive value accidentally matches the Sentry technical field value, but the chance of corruption becomes much lower.
As another change, it is worthwhile to configure Sentry logs in the following way:
They should send log messages of level WARNING and above to the regular Tribler logs so we have less chance of missing the case when sending the report to Sentry fails.
They should not spam Tribler logs with DEBUG-level messages.
Recently, I discovered that when an error occurs on my machine and the Feedback dialog suggests sending the report to the Tribler developer, clicking the "SEND REPORT" button does nothing. The confirmation box with the message "Successfully sent the report! Thanks for your contribution" shows, but no errors appear on the Sentry server.
Specifying the
debug=Trueoption when callingsentry_sdk.init(...)allows me to see Sentry logs configured weirdly:debug=Trueoption but too verbose. Without thedebug=Trueoption, even error-level log records from the Sentry are silenced.stderr, duplicating the same log messages instdoutandstderr. The formatting of these log records in thestderroutput was inconsistent with Tribler logs.With enabled Sentry logs, the following log record appeared in stderr:
It allowed me to investigate the reason for the error. As it turns out,
SentryScrubberproduced a corrupted error report that was then rejected by the Sentry server.SentryScrubbertries to remove sensitive information from the error report. It generates a set of sensitive strings with corresponding pseudo-random substitutes. For example, a username"John"may be replaced with the string"<foo>", and a machine name"JOHN-PC"may be replaced with the string"<bar>". Then, using regular expressions,SentryScrubberreplaces corresponding substrings in all places of the error report.As it turns out, the reason for the problem was that sometimes
SentryReporteddecides to replace an empty string""with the substitution"<>"and applies this replacement to all possible substrings. As a result, even elements added to the report by Sentry itself were transformed; for example, the identifier valuewas transformed into
and all other string elements of the report were changed in the same way.
As a simple fix,
SentryScrubbershould stop considering empty strings as a potential target for replacement. After such a fix, with the current logic,SentryScrubberstill has a chance to corrupt the report if some sensitive value accidentally matches the Sentry technical field value, but the chance of corruption becomes much lower.As another change, it is worthwhile to configure Sentry logs in the following way: