The deadlock was caused in transfermap.go by calling mu.RLock() in one
function then calling it again in a sub function. Normally this is
fine, however this leaves a window where mu.Lock() can be called. When
mu.Lock() is called it doesn't allow the second mu.RLock() and
deadlocks.
Thead 1 Thread 2
String():mu.RLock()
del():mu.Lock()
sortedSlice():mu.RLock() - DEADLOCK
Lesson learnt: don't try using locks recursively ever!
This patch fixes the problem by removing the second mu.RLock(). This
was done by factoring the code that was calling it into the
transfermap.go file so all the locking can be seen at once which was
ultimately the cause of the problem - the code which used the locks
was too far away from the rest of the code using the lock.
This problem was introduced in:
bfa5715017 fs/accounting: sort transfers by start time
Which hasn't been released in a stable version yet
This is preparation for getting the Accounting to check the context,
buf first we need to get it in place. Since this is one of those
changes that makes lots of noise, this is in a seperate commit.
During a copy/sync command, if an operation fails due to a network
issue and is retried, the underlying io.Reader is re-initialised,
but the stats for bytes already read are not reset, leading to incorrect
stats. THis was fixed by resetting the bytes read when an Account is
re-initialized.
Before this change we checked the transfer was out of range only
before the Read call. This means that we returned all the data to the
reader before declaring an error. This means that some backends wrote
the file even though an error was returned.
This fix checks the transfer after the Read as well, and chops the
excess characters off the read data if we are over the limit so that
we don't ever deliver all the data.
This fixes the tests introduced as part of 6f1766dd9e and #2672
on backends other than local.
Before this change the exit code for transfer limit exceeded was
incorrect. This was because the `resolveExitCode` function unwraps the
error thus reading the underlying error which is not the same as the
error it was comparing to (`ErrorMaxTransferLimitReached`).
This change fixes it by splitting the error definition in two so that
when the Fatal error is unwrapped we match against
`ErrorMaxTransferLimitReached` however when we return the error we
return `ErrorMaxTransferLimitReachedFatal`.
Before this change if an operation was retried on operations.Copy and
the operation was large enough to use an async buffer then an async
buffer was leaked on the retry. This leaked memory, a file handle and
a go routine.
After this change if Account.WithBuffer is called and there is already
a buffer, then a new one won't be allocated.
In 53a1a0e3ef we introduced a problem where if there was an
error on the file being transferred then the file was re-opened and
the old one wasn't closed.
This was partially fixed in bfbddab46b however this didn't
address the case of the old file being closed.
This is now fixed by
- marking the file as open again in UpdateReader
- moving the stopping the accounting machinery to a new method Done
core/stats can return two different schemas in 'transferring' field.
One is object with fields the other is just plain string.
This is confusing, unnecessary and makes defining response schema
more difficult. It also returns `lastError` as value which can be
rendered differently depending on source of error.
This change standardizes 'transferring' filed to always return
object but with reduced fields if they are not available.
Former string item is converted to {name:remote_name} object.
'lastError' is forced to be a string as in some cases it can be encoded
as an object.
This is done to make clear ownership over accounting object and prepare
for removing global stats object.
Stats elapsed time calculation has been altered to account for actual
transfer time instead of stats creation time.
This fixes several things wrong with the layout of the stats.
Transfers which haven't started are printed in the same format as
those which have so the stats with `--progress` don't show horrible
artifacts.
Checkers and transfers now get a ": checkers" and ": transfers" label
on the end of the stats line. Transfers will have the transfer stats
when the transfer has started instead of this.
There was a bug in the routine which shortened the file names (it
always produces strings 1 too long). This is now fixed with a test.
The formatting string was wrong with a fixed width of 45 - this is now
replaces with the value of `--stats-file-name-length`.
This also meant that there were unecessary leading spaces in the file
names. So the default `--stats-file-name-length` was raised to 45
from 40.
Before this change the moving average for the individual file stats
would start at 0 and only converge to the correct value over 15-30
seconds.
This change starts the weighting period as 1 and moves it up once per
sample which gets the average to a better value instantly.
Previous to this change package used for this
github.com/VividCortex/ewma took a 0 average to mean reset the
statistics. This happens quite often when transferring files though a
buffer.
Replace that implementation with a simple home grown one (with about
the same constant), without that feature.
This removes the old system of part accounting and replaces it with a
system of popping off the accounting reader and wrapping up new ones
as necessary.
This makes it much easier to carry the context down the chain of
wrapped readers and get the limiting as near as possible to the
output. This makes the accounting more accurate and the bandwidth
limiting smoother.
Fixes#2029 and Fixes#1443
The purpose of this is to make it easier to maintain and eventually to
allow the rclone backends to be re-used in other projects without
having to use the rclone configuration system.
The new code layout is documented in CONTRIBUTING.
