Part of overall effort to clean up, unify, use link monitoring more,
and make Tailscale quieter when all networks are down. This is especially
bad on macOS where we can get killed for not being polite it seems.
(But we should be polite in any case)
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
+ add a test for parseAndRemoveLogLevel()
+ add a test for drainPendingMessages()
+ test JSON log encoding including several special cases
Other tests frequently send logs but a) don't check the result and
b) do so by happenstance, such that the code in encode() was not
consistently being exercised and leading to spurious changes in
code coverage. These tests attempt to more systematically test
the logging function.
This is the second attempt to add these tests, the first attempt
(in https://github.com/tailscale/tailscale/pull/1114) had two issues:
1. httptest.NewServer creates multiple goroutine handlers, and
logtail uses goroutines to upload, but the first version had no
locking in the server to guard this.
Moved data handling into channels to get synchronization.
2. The channel to notify the test of the arrival of data had a depth
of 1, in cases where the Logger sent multiple uploads it would
block the server.
This resulted in the first iteration of these tests being flaky,
and we reverted it.
This new version of the tests has passed with
go test -race -count=10000
and seems solid.
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
This reverts commit e4f53e9b6f.
At least two of these tests are flakey, reverting until they can be
made more robust.
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
* logtail: test parseAndRemoveLogLevel()
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
* logtail: test JSON log encoding.
Expand TestUploadMessages to also exercise the encoding functions
in logtail, like JSON logging and timestamps.
Other tests frequently send logs but a) don't check the result and
b) do so by happenstance, such that the lines in encode() were not
consistently being exercised and leading to spurious changes in
code coverage.
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
* logtail: add a test for drainPendingMessages
Make the client buffer some messages before the upload server
becomes available.
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
* logtail: use %q, raw strings, and io.WriteString
%q escapes binary characters for us.
raw strings avoid so much backslash escaping
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
Right now TestFastShutdown tries to upload logs to localhost:1234,
which will most likely respond with an error. However if one has an
actual service running on port 1234, it would receive a connection
attempting to POST every time the unit test runs.
Start a local server and direct the upload there instead.
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
Start an HTTP server to accept POST requests, and upload some logs to
it. Check that uploaded logs were received.
Code in logtail:drainPending was not being reliably exercised by other
tests. This shows up in code coverage reports, as lines of code in
drainPending are alternately added and subtracted from code coverage.
This test will reliably exercise and verify this code.
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
Log levels can now be specified with "[v1] " or "[v2] " substrings
that are then stripped and filtered at the final logger. This follows
our existing "[unexpected]" etc convention and doesn't require a
wholesale reworking of our logging at the moment.
cmd/tailscaled then gets a new --verbose=N flag to take a log level
that controls what gets logged to stderr (and thus systemd, syslog,
etc). Logtail is unaffected by --verbose.
This commit doesn't add annotations to any existing log prints. That
is in the next commit.
Updates #924
Updates #282
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add content length hints to headers.
The server can use these hints to more efficiently select buffers.
Stop attempting to compress tiny requests.
The bandwidth savings are negligible (and sometimes negative!),
and it makes extra work for the server.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
Also, bit of behavior change: on non-nil err but expired context,
don't reset the consecutive failure count. I don't think the old
behavior was intentional.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
We want to run bo.Backoff() after every upload, regardless. If
upload==true but err!=nil, we weren't backing off, which caused some
very-high-throughput log upload retries in bad network conditions.
Updates #282.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
If a test calls log.Printf, 'go test' horrifyingly rearranges the
output to no longer be in chronological order, which makes debugging
virtually impossible. Let's stop that from happening by making
log.Printf panic if called from any module, no matter how deep, during
tests.
This required us to change the default error handler in at least one
http.Server, as well as plumbing a bunch of logf functions around,
especially in magicsock and wgengine, but also in logtail and backoff.
To add insult to injury, 'go test' also rearranges the output when a
parent test has multiple sub-tests (all the sub-test's t.Logf is always
printed after all the parent tests t.Logf), so we need to screw around
with a special Logf that can point at the "current" t (current_t.Logf)
in some places. Probably our entire way of using subtests is wrong,
since 'go test' would probably like to run them all in parallel if you
called t.Parallel(), but it definitely can't because the're all
manipulating the shared state created by the parent test. They should
probably all be separate toplevel tests instead, with common
setup/teardown logic. But that's a job for another time.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Some programs use frequent short-duration backoffs even under non-error
conditions. They can set this to avoid logging short backoffs when
things are operating normally, but still get messages when longer
backoffs kick in.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We'll be fixing the server so this won't trigger in practice,
but it demos the connection reuse problem.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Its semantics has changed slightly, this will let us use it to
drive batched logging in special circumstances.
Signed-off-by: David Crawshaw <crawshaw@tailscale.com>
In particular, the Dup2 syscall is not defined in the syscall
package on arm64, but is defined in x/sys/unix.
Signed-off-by: David Anderson <dave@natulte.net>
