RBE has signifigant performance overhead during
analysis time, in theory to reduce execution time.
But multiproduct_kati only tests analysis, and does
not run any execution, so RBE is just slowing us
down.
Bug: 229863781
Bug: 265603584
Test: time ./build/soong/build_test.bash --products aosp_arm64 on stage_aosp_master
Change-Id: Icbb2f001fee03a75bad37c14d08a203be20ac6cf
multiproduct_kati assumed each build would take ~30GB of RAM,
but it seems it's closer to 40GB now.
The current build_test builds use 7 jobs, so this would bring
us down to 5, which should still keep us under the 6 hour
build time limit as currently these tests are at ~3 hours.
Bug: 262629589
Test: Presubmits
Change-Id: I51af364cc602a3bf7ba85fd8f92c4dd57585abbd
These are needed to add staging-mode to some throttled builds. See go/roboleaf-launch-tests for details.
Test: m nothing
Test: build/soong/build_test.bash --bazel-mode-staging
Change-Id: I456ef80fb9c50579a28fba3dbd184d4e4653dde7
Device builds are no longer supported on Mac, but we do support building
various host tools, including the SDK build-tools and platform-tools
packages. These have dependencies on [java] device modules, so we don't
completely disable device modules, only hide them from Make (which makes
them more difficult to trigger from the command line).
Also fix the mac build of multiproduct_kati, so that `m blueprint_tools`
works on Mac.
Bug: 187222815
Test: `m`, `m dist`, etc on Mac
Change-Id: I92f16605d5cd173d431cbcb79081234d45cc6e2e
If SOONG_UI_ANSI_OUTPUT is set to a true value force the simple status
output to keep ANSI codes. This will allow buildbots to collect logs
with ANSI codes in them. Smart status output is not affected as it
always keeps ANSI codes.
Bug: 147310922
Test: manual
Change-Id: I0b78ceebb65125b8e8dafb4787816fb679d3eb3e
I erroneously assumed that the continuous builds don't set DIST_DIR,
whereas they do.
The variable naming is quite questionable, but for now, the important
thing is that the code works and I wanted to do that with the least
amount of change possible.
Also add some more logging:
- Shell environment
- Command line arguments
- Location of logs.zip
- The logs directory
Test: Presubmits.
Change-Id: I651b1045eeef568aa860fd2457fa7799ad68ba38
This is done by moving SetupSignals() to its own little package.
There are a number of tiny little utility packages for soong_ui we might
be better of merging, but that's for another change (maybe)
Test: Presubmits.
Change-Id: I07b0ca98bfb8884ef4223d665e632183b9896a0d
This mostly cuts the dependency from multiproduct_kati to the build
package; the remaining logic is SetupSignals which will be dealt with in
a followup change.
This requires some code duplication (e.g. ensureFileExists() and
detectTotalRAM()).
This removes a Finder invocation added in aosp/452598; it looks like
it's not a measurable performance hit. The quoted half a second does not
matter much in build_test{1,2,3,4} which take an amount of time measured
in hours.
It also makes ANDROID_QUIET_BUILD unsupported for the UI of
multiproduct_kati; the function of that variable seems to be to suppress
the output from build actions, which don't exist in multiproduct_kati so
it should not matter. In addition, I checked whether anyone sets this
flag for multiproduct_kati and it doesn't appear to be the case.
Test: Presubmits.
Change-Id: I1c2901e1ea69c3a35429e6c967a7bba845676ef7
This serves to not link parts of soong_ui (and eventually soong_build)
into a separate, weird binary. This is in turn good because they contain
any number of global variables and no one really thought about what
happens when two instances are executing at the same time in the same
address space.
This comes with a slight performance hit: 5 aosp_* projects build 152
seconds instead of 146. I suppose this is a price worth paying for a
clean design?
Test: presubmits.
Change-Id: I5623dcab2290f0fc392dd2ede597b9794a3d2a4e
- Rename the "BuildX" variables to "RunX"
- Remove redundant comments
- Inline all the "what to do" based on config in build.go
- Inline some constants only used in one place
Bug: 189187214
Test: m nothing
Test: build/soong/build_test.bash
Change-Id: I111a69e642212d7938d4971283545e0d9acbb01a
Merged-In: I111a69e642212d7938d4971283545e0d9acbb01a
(cherry picked from commit d274ea9196)
I6dca478f356f56a8aee1e457d71439272351390b replaced calling
build/blueprint/bootstrap.bash with running minibp directly,
but didn't propagate the EMPTY_NINJA_FILE environment variable
to minibp. Since everything that uses EMPTY_NINJA_FILE is
executed directly from soong_ui now, replace the EMPTY_NINJA_FILE
environment variable with a config bool and pass it to minibp.
Bug: 189148777
Test: build/soong/build_test.bash --dist --incremental --shard-count=300 --shard=1 && du out/multiproduct
Change-Id: I4d64275ce02c5d68948012f71ac4dc3795af9e85
multiproduct_kati sometimes passes 20GB per soong_build process,
increase required ram per job to 25GB.
Bug: 178376498
Test: none
Change-Id: I446f52ea7419bc3015d940511af1f03b15eb651e
Also get Bazel to write real files there (not symlinks) so that the DIST_DIR can be independent.
Test: Manually using e.g. DIST_DIR=/tmp/foo USE_BAZEL=1 m dist
Change-Id: I39d5219500864c9ecc85f356a028e9b5bf2607f4
There is no need for multiproduct_kati to run the tests for
every product, they don't vary by product config. --skip-soong-tests
can also be used for local development to run soong_build even if
the tests don't pass.
Bug: 156428456
Test: m --skip-soong-tests nothing
Change-Id: I9c00e3d1b6e51d17bb290339c3f124d4d1c9e69f
This reverts commit 323dc60712.
Reason for revert: Possible cause of test instability
Bug: 170513220
Test: soong tests
Change-Id: Iee168e9fbb4210569e6cffcc23e60d111403abb8
Putting t.Parallel() in each test makes them run in parallel.
Additional t.Parallel() could be added to each subtest, although
that requires making a local copy of the loop variable for
table driven tests.
Test: m checkbuild
Change-Id: I5d9869ead441093f4d7c5757f2447385333a95a4
Apparently we have some instances of Soong taking 16GB of RAM. While
that's a problem to solve on it's own, make multiproduct_kati's
auto-parallelism detection smarter by taking into account the total RAM
on the machine instead of just the number of CPUs.
Bug: 146925549
Test: check soong.log for autodetected parallelism values
Change-Id: Ice34c2cf73ea4f8f235521cbefc8654a79a04eef
This is so that we can split the build_test build on large branches into
multiple builds, each testing a subset of the products.
Test: m blueprint_tools (runs the go unit tests)
Test: build/soong/build_test.bash --only-config
vs
Test: build/soong/build_test.bash --only-config --shard-count=2
Test: build/soong/build_test.bash --only-config --shard-count=2 --shard=2
Change-Id: I40ccc1aa477bc0ffa74ff564d155068509be18f0
Force soong_ui in dumpvars mode to use a dumb terminal so that it
doesn't print an empty status table. Also move the logs to files
that are prefixed with "dumpvars-" so that they don't rotate the
real build logs.
Fixes: 139307849
Test: lunch
Change-Id: If92421af07731fedc9ca42670f309e95922a2224
Write log output through StatusOutput so that the status implementation
can synchronize it with its own output.
Test: status_test.go
Change-Id: I917bdeeea4759a12b6b4aa6d6d86ee18a2771723
terminal.Writer is now just a wrapper around stdio.Stdout() without
any useful functionality. Replace it with stdio.Stdout() as an
io.Writer.
Test: ui/terminal/status_test.go
Change-Id: I5bc5476afdca950b505642f0135a3af9d37fbe24
If ANDROID_QUIET_BUILD environment variable is set to '1', do not show 20
lines of the environment variables. In addition, for the failing step show
only the step name and output, omitting the command proper (the verbose log
as well as error.log still contains it).
If build succeeds, the output of the build itself consists of a single
line:
```
> m androidmk
[100% NN/NN] <last command>
```
When it fails, the output does not contain sometimes very long command
line:
```
> m androidmk
[ 97% NN/MM] test androidmk
FAILED: <step>
--- FAIL: TestEndToEnd (0.01s)
androidmk_test.go:1025: failed testcase 'prebuilt_etc_TARGET_OUT_ETC'
input:
include $(CLEAR_VARS)
LOCAL_MODULE := etc.test1
LOCAL_MODULE_CLASS := ETC
LOCAL_MODULE_PATH := $(TARGET_OUT_ETC)/foo/bar
include $(BUILD_PREBUILT)
expected:
prebuilt_etc {
name: "etc.test1",
filename: "foo/bar",
}
got:
prebuilt_etc {
name: "etc.test1",
filename: "foo/bar",
}
FAIL
17:50:53 ninja failed with: exit status 1
```
[The related change in build/make/envsetup.sh suppresses timing display
when the same variable is set.]
Change-Id: I4d3c72457de031ff58a324c2fe98f4c1d10f8239
Test: treehugger
Stop writing out ninja files, as they become very large, especially when
multiplied by the number of defined products.
Test: treehugger
Change-Id: Id9529d14040acb72a0188e58b5db2911f142071e
Make the cleanspec & Android.mk runs of kati use much of the same code
and arguments.
Also renames 'Kati' to 'KatiBuild' in many cases, in the the expectation
that we'll have a 'KatiPackage' step in the future.
Use --no_ninja_prelude and move local_pool & _kati_always_build_ into
the combined ninja file. This will reduce the need to re-read makefiles
when Goma is enabled, and it allows us to include more than one
Kati-generated ninja file in the build graph.
Bug: 116968624
Test: build_test on downstream branches
Change-Id: Ibdac689b81f62dc293647fad917d84946f2c3cfa
Add test that cover basic command line usage of soong_zip. -D
is not covered yet as the implementation will be replaced with
one that is also more easily testable in the next patch.
Bug: 116751500
Test: zip_test.go
Change-Id: I5a1bcee74ebc9cb3cf332c36f89bc12c0e807ad2
Move the actual per-product builds out of the main function, combining
the product configuration and build sections. This makes it a lot more
readable.
This ends up saving a lot of FDs -- a -only-config build on aosp with 77
products used to require ~500 FDs, it succeeds with a limit of 150 FDs
now. I'm leaving the code to bump our FD limit though, since 2x the
number of internal products is too close to the 1024 limit for comfort.
Bug: 70370883
Test: prlimit -n150:150 build/soong/build_test.bash -only-config
Change-Id: Ia559beadc19deb8a5b9d50af6e0276e846fd8608
This uses a very large amount of intermediates, but can speed up
subsequent multiproduct_kati runs by not re-reading all the makefiles
unless necessary.
Since we'll no longer have the full build output on every run,
explicitly save away a std_full.log of the log from the last time we ran
Kati.
Test: build/soong/build_test.bash -products aosp_arm -incremental
Change-Id: Iaae580c0e49a2024528617d39b2b477e29d99250
Allows specifying make goals when using multiproduct_kati, for example:
$ multiproduct_kati -only-config dump-products
This makes dumping products with multiproduct_kati easier than before.
Test: multiproduct_kati -only-config -products=full,aosp_arm64 -out=/tmp/p dump-products
Change-Id: If526bd2ba1b7bbadcb230c020202ef9e421e1852
This adds a new status package that merges the running of "actions"
(ninja calls them edges) of multiple tools into one view of the current
state, and gives that to a number of different outputs.
For inputs:
Kati's output parser has been rewritten (and moved) to map onto the
StartAction/FinishAction API. A byproduct of this is that the build
servers should be able to extract errors from Kati better, since they
look like the errors that Ninja used to write.
Ninja is no longer directly connected to the terminal, but its output is
read via the protobuf frontend API, so it's just another tool whose
output becomes merged together.
multiproduct_kati loses its custom status routines, and uses the common
one instead.
For outputs:
The primary output is the ui/terminal.Status type, which along with
ui/terminal.Writer now controls everything about the terminal output.
Today, this doesn't really change any behaviors, but having all terminal
output going through here allows a more complicated (multi-line / full
window) status display in the future.
The tracer acts as an output of the status package, tracing all the
action start / finish events. This replaces reading the .ninja_log file,
so it now properly handles multiple output files from a single action.
A new rotated log file (out/error.log, or out/dist/logs/error.log) just
contains a description of all of the errors that happened during the
current build.
Another new compressed and rotated log file (out/verbose.log.gz, or
out/dist/logs/verbose.log.gz) contains the full verbose (showcommands)
log of every execution run by the build. Since this is now written on
every build, the showcommands argument is now ignored -- if you want to
get the commands run, look at the log file after the build.
Test: m
Test: <built-in tests>
Test: NINJA_ARGS="-t list" m
Test: check the build.trace.gz
Test: check the new log files
Change-Id: If1d8994890d43ef68f65aa10ddd8e6e06dc7013a
When the config loading was changed to using AndroidProducts.mk lists
from soong_ui instead of finding them itself, multiproduct_kati stopped
being able to find all the products. Move the finder initialization
earlier so that we can dump the full list.
Test: build/soong/build_test.bash -only-config
Test: forrest runs of every downstream build_test
Change-Id: I085d40f9008a47b850783499365977a0ff70ac66
On large branches (250+ configurations), the open file count can go
over the default 1024 soft limit on Ubuntu. Many systems have increased
that default, but for the ones that haven't, at least opt into the
hard limit until this can be refactored to use fewer open files.
Bug: 70370883
Test: prlimit -n256:4096 build/soong/build_test.bash -only-config
Test: ulimit -Sn 256; build/soong/build_test.bash -only-config (darwin)
Change-Id: I7a952ffc89a0149ab65b04db1523d348daa6ab3e
Instead of throwing it away (though it would also be in soong.log along
with the verbose messages).
Test: build/soong/build_test.bash with ctx.Println added
Change-Id: I64f8b11ab6752a6be8a5934fb41e30439238e331
Instead of deleting artifacts/logs from successful build (unless -keep
is set), and keeping unsuccessful artifacts, keep all logs and remove
all artifacts (unless -keep is set, then we'll compress the artifacts).
If -dist is passed in, we'll put an archive of the logs into the
DIST_DIR. Even compressed, the rest of the artifacts are still a bit too
large to dist (~5.6GB on AOSP).
Test: build/soong/build_test.bash
Test: build/soong/build_test.bash -keep
Test: build/soong/build_test.bash -dist
Change-Id: I87f55978c18c8ff2e517b8271554ba383003742f
This speeds up dumping make variables from ~380ms using make to ~220ms
using ckati. It also means that we're consistently using the same parser
for builds (with the same .KATI_READONLY/etc extensions).
envsetup.sh (lunch) / other scripts still use make, changing those to go
through soong_ui will be a future change.
Test: m clean; m nothing
Test: USE_GOMA=true m nothing
Test: m PRODUCT-aosp_x86-sdk
Test: m APP-Calculator
Test: build/soong/build_test.bash -only-config (on AOSP and internal master)
Change-Id: I6ca554de8de4955fb869001d06d29969b75751cc
This can make it easier to add build/soong/build_test.sh to continuous
build branches to catch new breaks, but where some products are known to
be broken.
Test: build/soong/build_test.bash --skip-products aosp_arm,aosp_arm64
Change-Id: Ib240570b68957077047ca75b78658429cb8c6a49
The Finder runs roughly 200ms faster than findleaves.py in aosp,
and runs roughly 400ms faster in internal master.
Bug: 64363847
Test: m -j
Change-Id: I62db8dacc90871e913576fe2443021fb1749a483
Adds a -dist argument that will use DIST_DIR to save logs.
Also writes a summary of each std.log to stderr on errors, so that the
error is more likely to show up in the error reporting. This output is
prefixed with "> " to differentiate it from the progress reports from
multiproduct_kati itself.
Test: multiproduct_kati -only-config
Test: DIST_DIR=dist build/soong/build_test.bash -dist (introducing errors)
Change-Id: I5005b5f3f200c876bc004dd9b0e01e7b6edf5be2
It now uses the same output style as ninja, overwriting status lines in
smart terminals.
Test: multiproduct_kati
Test: multiproduct_kati | cat
Change-Id: I8db5198ffdc5ebc5503241ac492379753d92978e
