Similarly to the codeparser change, change to sha256 hashes due to worries
over collisions. The main impact of this change is slightly slower parsing
time as well as longer sstate file names.
(Bitbake rev: 66f1b766997d53b4375fdd25719b1175f3828903)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We've reports of hash collision with codeparser. Looking at the way collision
problems occur with md5 and the way our function templating works, I can believe
we may run into issues.
This patch therefore switches to sha256.
Performance wise, parse time could appear to rise by 4s in 374s
Before:
384329 in 2.966s (md5)
After:
349743 in 2.340s (sha256)
34723 in 1.245s (md5)
since we still have md5 used elsewhere in the code, something we should look at
next (using sha256 everywhere is around 5.3s in total)
Unfortunately this does nearly double the size of the codeparser cache file
due to the hash size change.
(Bitbake rev: 4bed7a97198176503fe8c72d8dd7c36b28fc9dd1)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We have two choices, split the recipes amongst the parsing threads in
blocks ahead of time, or have a queue which parsers pull from when idle.
The optimum approach depends on how similar the pieces are. For the single
recipe reparse case, there is currently a significant wait for the feeder
thread to start (around 0.25s in a 2s command).
Its possible splitting into blocks in advance may be unluckly for some other
workloads but experimentally it seems to work better overall for me at least.
(Bitbake rev: ae79868861568d673a70472e85a4bde9e2d84a8f)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There used to be many bugs in multiprocessing and we implemented our own
feeder process to avoid them. Now that we have python 3.x, these are fixed
and just using the standard Queue mechanism appears to work fine. We can
therefore drop the unneeded code and simplify.
(Bitbake rev: b2d39fc37fcf3c81a562ec1ef4f8b4c1a493fc57)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add a test for x86_64 in overrides where is was being incorrectly handled.
There was a previous fix (3a3be518536acc868c7eeb3c1111ad1b321480b7) but
this ensures we don't regress.
(Bitbake rev: 08c314eac231ac9292c8c95f1c5a6fc2023fe749)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Under heavy load with process delays, sqlite can issues a "locking
protocol" error (SQLITE_PROTOCOL). Unfortunately, it is impossible to
distinguish between actual locking protocol errors and this race
condition, so they best that can be done is to retry the operation when
the error is detected.
[YOCTO #13108]
(Bitbake rev: 93cd15644f9d12b38abea276fee7b5bade0276df)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If the unique hash is being used to track task dependencies, the hash
validation function needs to know about it in order to properly validate
the hash.
[YOCTO #13030]
(Bitbake rev: 9a529bb2658a4046dafbf32e1eb503d84e64e947)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The unique hash is now passed to the task in the BB_UNIHASH variable
[YOCTO #13030]
(Bitbake rev: aab80b099f6f259e4b57cba2c26dd385d07c5947)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Requests the task unique hash from siggen and tracks it
[YOCTO #13030]
(Bitbake rev: 1ecc47f0831b35c8c92b37a81cef4e43ff9f67b2)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Abstracts the function to get the unique hash for a task. This hash is
used as in place of the taskhash for the purpose of determine how other
tasks depend on this one. Unless overridden, the taskhash is the same as
the unique hash, preserving the original behavior.
[YOCTO #13030]
(Bitbake rev: d0065b34cea81fcadff14b0383779e9559d94508)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When an indirect multiconfig dependency exists, such as:
A depends on B, B has a multiconfig dependency to C,and our build
target is A, the multiconfig dependency to C is not processed on
time, hence no providers are added for it, causing an exception in
the runqueue because the dependency does exist in it.
Call add_unresolved() for all available multiconfigs before processing
providers for multiconfig dependencies, detecting mcdepends on time so
providers for them can be added correctly.
(Bitbake rev: 8a6bc7584ad61b4de98af92a86066602006262f9)
Signed-off-by: Alejandro Enedino Hernandez Samaniego <alejandr@xilinx.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If the bitbake server recieved multiple connections, it currently closes
ones it can't handle (while its dealing with another). This is rather
antisocial behaviour which causes clients to quickly run through their
retries and abort.
Instead, queue any other connections until the current one is closed. This
way the client can decide when it wants to stop waiting for the server. If the
client is gone by the time we handle it, we handle that gracefully.
This also fixes a number of bugs in the connection handling where connections
which did drop early were badly handled causing tracebacks in the logs.
Also, handle queue incomming connections in a loop to ensure that the main
client handling doesn't starve that piece of the system.
This code was stress tested by running 50 connection attempts in parallel at
once, ensuring the code correctly handled them.
(Bitbake rev: 02845a561b38658ac3edf5cc9c34625ed860d34f)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Now that we set a timeout for the socket, it can return EWOULDBLOCK
if a signal or other event happens to wake up even if we don't timeout.
If this happens, retry the connection, else we simply see it quickly
loop through the retries and abort the connection in a very short
interval.
(Bitbake rev: d5b0a9a302ac0eafa4f797ac15ea77db87e82b3c)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The current bitbake output makes it hard to know which retry is being
attempted. Add this information to the output to make it clearer.
(Bitbake rev: 0774e6e03d27adb7aca6fa9c47ab6ad426c937de)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The current value of 2 seconds has shown to be short in
wider testing.
(Bitbake rev: 8a1f2fcf35d61d83bbafa8fa3ae215fd5f51728b)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Adds a test suite for testing the persistent data cache
[YOCTO #13030]
(Bitbake rev: 96a4155049e834af17069d981cc2215e50d18c1a)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We're seeing hangs in oe-selftest where server startup and shutdown are
racing. The assumption was a connect would timeout however no timeout is
set which can leave processes hanging. Set a short timeout for
the connection to avoid this.
(Bitbake rev: f02114cb70e8f6f1d32e19c02b758fe0aadecd19)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
With aarch64 hosts coming into use, set the syscall number to
avoid ioprio warnings on that platform.
(Bitbake rev: 5eaf9e7b26f09f5f106e1c3c6976d517b289450a)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We're seeing problems due to the way x86-64 is handled (or not handled)
as an override. Relax the containts on overrides from being lowercase
to being lowercase or numeric. This fixes problem where MACHINE=qemux86
would work but MACHINE=qemux86-64 would fail the same tests.
(Bitbake rev: 3a3be518536acc868c7eeb3c1111ad1b321480b7)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Fix some further python3 warnings about unescaped regexs.
(Bitbake rev: 8667605d016e82add95638fcb15c2bbc1b489ecc)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The p_error() is used for printing errors when parse shell code, but it can't
the EOF error correctly
- Add the following lines to quilt.inc
do_configure_prepend () {
find ${s} -name "*.in" -exec sed -i -e "1s,^#\!.*@perl@ -w$,#\! @perl@\nuse warnings;," {} \;
if [ hello ]; then
}
- Before the patch:
$ rm -fr cache/ tmp/cache/; bitbake -p
[snip]
WARNING: /path/to/quilt/quilt-native_0.65.bb: Error during finalise of /path/to/quilt/quilt-native_0.65.bb
[snip]
bb.pysh.sherrors.ShellSyntaxError: None
followed by:
We can see that this isn't easy to debug, let p_error() check wheter it is EOF
and print appropriate errors can improve the error message. And don't let
codeparser.py except pyshlex.NeedMore (in fact, it never worked since p_error()
only raise ShellSyntaxError), but make it print the last 5 lines which might be
useful for debuging.
- After the patch
$ rm -fr cache/ tmp/cache/; bitbake -p
[snip]
ERROR: /path/to/quilt/quilt_0.65.bb: Error during parse shell code, the last 5 lines are:
find /path/to/quilt/0.65-r0/quilt-0.65 -name "*.in" -exec sed -i -e "1s,^#\!.*@PERL@ -w$,#\! @PERL@\nuse warnings;," {} \;
if [ hello ]; then
autotools_do_configure
sed -e 's,^COMPAT_SYMLINKS.*:=.*,COMPAT_SYMLINKS :=,' -i /path/to/quilt/0.65-r0/quilt-0.65/Makefile
[snip]
File "/path/to/bb/pysh/pyshyacc.py", line 649, in p_error(p=None):
w('Unexpected EOF')
> raise sherrors.ShellSyntaxError(''.join(msg))
bb.pysh.sherrors.ShellSyntaxError: Unexpected EOF
(Bitbake rev: 44790597951638e32eb1672de2e40bd5a603326b)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
>>> from Set import Set
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ImportError: No module named 'Set'
So remove related code.
(Bitbake rev: 118e9d8894e64d102543562314b318676006e710)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The only two that we need are ShellSyntaxError and ShellError, others are not
used, so remove them.
(Bitbake rev: abee2ff5ee71bf8c3221cd9d3cd4243ebb4ca273)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The configuration of the sqlite database can timeout due to locking
under heavy load and should be subject to the same retry logic as the
other statements.
[YOCTO #13069]
(Bitbake rev: 5a2a95b0396e39662968690b3065d2f88167a71c)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We've already tweaked remove() for speed and not to error if it
races for deletion. Therefore use this for prunedir() which was
starting to show the same bug reports.
[YOCTO #13003]
(Bitbake rev: 47f359f9b017f8d99d0bb2161ac0dcefcbd915de)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
os.errno used to happen to work but is invalid. Correct to use errno.
[YOCTO #13068]
(Bitbake rev: 03aae96a8859409b1ce7e3d5c1197371fa96bd14)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This could clearly never have worked since the python3 migration
but as its in an error path, it doesn't get tested/used much.
(Bitbake rev: 704f27dc28d50a6dc02c8b64274ee4ecb3058c4a)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We're seeing slow startup in bitbake, add some timeing debug messages so
the logs are more useful for debugging when its slow.
(Bitbake rev: 8d1fc115b8a176009f1f3a8ce840b422e7e0b45e)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
On loaded production systems we've seen bitbake server take over
40s to start up. Increase the timeout to 90s which tries to avoid
failures in selftests.
The delays come from setting up the inotify watches (31s) so can't
really be avoided.
After 5s delay we now warn the user we're waiting for 90s so the
interactive exeperience shouldn't be much changed and its very
unlikely the user would see that anyway.
(Bitbake rev: 7a4e105093c3080c1087cd06abc2883852a23e28)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Remove a deprecated warning and stop using our own deprecated API!
(Bitbake rev: 83ece2c6f4b000e906fec9148f25bd1dff66cfb0)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The mechanism used to get the hash for a stamp file is split out so that
it can be overridden by derived classes
[YOCTO #13030]
(Bitbake rev: ce241534d19b2f1c51dbdb3b92419676d234e464)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Pass the task hash as a parameter to the 'runtask' message instead of
passing the entire dictionary of hashes when the worker is setup. This
is possible less efficient, but prevents the worker taskhashes from
being out of sync with the runqueue in the event that the taskhashes in
the runqueue change.
[YOCTO #13030]
(Bitbake rev: 1e86d8c1bec7ea5d016a5ad2097f999362e29033)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Turns off the shared cache. It isn't a significant factor in performance
(now that WAL is enabled), and is a really bad idea to have enabled in
processes that fork() (as bitbake it prone to do).
[YOCTO #13030]
(Bitbake rev: 71b427bf01374973a971c10cb64024c8ef2a11eb)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Enabling the write ahead log improves database reliability, speeds up
writes (since they mostly happen sequentially), and speeds up readers
(since they are no longer blocked by most write operations). The
persistent database is very read heavy, so the auto-checkpoint size is
reduced from the default (usually 1000) to 100 so that reads remain
fast.
[YOCTO #13030]
(Bitbake rev: 79100fa67539f9654af9bf6d3e6842eb5c12e989)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Constructs the "key" column in the persistent database as a non-NULL
primary key. This significantly speeds up lookup operations in large
databases.
[YOCTO #13030]
(Bitbake rev: f5ba7775cfcb90401522d977cc66fe0f5aeb7a66)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The original implementation of persistent data executed all SQL
statements via sqlite3.Connection.execute(). Behind the scenes, this
function created a sqlite3 Cursor object, executed the statement, then
returned the cursor. However, the implementation did not account for
this and failed to close the cursor object when it was done. The cursor
would eventually be closed when the garbage collector got around to
destroying it. However, sqlite has a limit on the number of cursors that
can exist at any given time, and once this limit is reached it will
block a query to wait for a cursor to be destroyed. Under heavy database
queries, this can result in Python deadlocking with itself, since the
SQL query will block waiting for a free cursor, but Python can no longer
run garbage collection (as it is blocked) to free one.
This restructures the SQLTable class to use two decorators to aid in
performing actions correctly. The first decorator (@retry) wraps a
member function in the retry logic that automatically restarts the
function in the event that the database is locked.
The second decorator (@transaction) wraps the function so that it occurs
in a database transaction, which will automatically COMMIT the changes
on success and ROLLBACK on failure. This function additionally creates
an explicit cursor, passes it to the wrapped function, and cleans it up
when the function is finished.
Note that it is still possible to leak cursors when iterating. This is
much less frequent, but can still be mitigated by wrapping the iteration
in a `with` statement:
with db.iteritems() as it:
for (k, v) in it:
...
As a side effect, since most statements are wrapped in a transaction,
setting the isolation_level when the connection is created is no longer
necessary.
[YOCTO #13030]
(Bitbake rev: e8b9d3f534ef404780be23b601d5a4bb9cec928a)
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Small tweak to ensure these items are printed into the log since there
is other logging code which looks for the header and this makes it clearer
the server did start but is slow somewhere in startup.
(Bitbake rev: 7c0b84cac892744225fa0462f918ea9a79b356cb)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Its possible the read may return a smaller number of characters. Remove
the possibility by using a single character to signal the server is ready.
(Bitbake rev: 767c9596d129d21ddf5d2e00c55f0a0525c641dc)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This simplifies the code and makes it easier to read but has the
same functionality.
(Bitbake rev: a0b9cfaf2d03fd047a79d32e668001718d02c4bf)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The previous check was in data.py which only can check code like "python
funcname()" in the dependency chain, but there are 3 kinds of python functions:
- python()
- def py_funcname()
- python funcname()
Add the checking to BBHandler to check and warn for all of them.
The warning looks like:
WARNING: /path/to/recipes-core/busybox/busybox_1.29.2.bb: python should use 4 spaces indentation, but found tabs in busybox.inc, line 75
(Bitbake rev: 0cdc5b81fc1f5e5281a525a657e420ebc3bb9e90)
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The ${DL_DIR}/svn directory is used by BitBake to keep checked-out SVN
repositories from which tarballs are generated. These repositories were
protected from concurrent update with a lock on the tarballs. However,
the tarballs are specific to the SRCREV and module checked out (many
tarballs can come from the same repository), meaning a repository could
be modified concurrently if two recipes checked out two different
SRCREVs or modules from it in parallel. This caused errors like the
following:
ERROR: Fetcher failure: Fetch command failed with exit code 1, output:
svn: E155004: Run 'svn cleanup' to remove locks (type 'svn help cleanup' for details)
svn: E155004: Working copy '/home/foo/downloads/svn/repo/trunk' locked.
svn: E155004: '/home/foo/downloads/svn/repo/trunk' is already locked.
Fix it by adding a per-repository lock that's independent of the module
and SRCREV.
(Bitbake rev: 3f1f183a17bf3580da8a4ffd6dab30b62c2654a8)
Signed-off-by: Ulf Magnusson <Ulf.Magnusson@bmw.de>
Signed-off-by: Michael Ho <Michael.Ho@bmw.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Currently if there are no setscene tasks, the disk monitor isn't started.
Move the startup code to somewhere to ensure it always is started. This
issue would partially explain occasional selftest failures.
(Bitbake rev: 5ba83ee25c1c9cba349edb68a22476b1d5fca6ce)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
10 log lines may not capture any full traceback, increase the number of
lines to 60 which covers most tracebacks.
(Bitbake rev: b8538c2088c199157404712a60a6740c926ce312)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We're seeing issues where the server doesn't start with no logs as to why. Allow
the server to print the last 60 log lines just in case this shows us something useful
about what is failing.
(Bitbake rev: c8c80b404e38fe96f65d6314cd95f4069319f3d6)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Also use assetIn instead of assertTrue which aids debugging failures.
(Bitbake rev: 19dee675bb9ad012d28e1e57a888931355a831cb)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
