Networks will be properties of NicParams and NodeGroups. Therefore
they need to be defined before those.

Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

This patch removes code which is no longer used due to refactoring:

- http.InitSsl, last usage removed in commit 33231500 (“Convert RPC
  client to PycURL”)
- rapi.baserlib.MakeParamsDict, last usage remove in commit 4e5a68f8
  (“RAPI: rlib1 removal”)
- rapi.baserlib.ExtractField, last usage removed in commit 028c6b76
  (“RAPI:Fix root list and unittest for it.”)
- qa.qa_utils.AssertNotEqual, last real usage removed in commit
  2f4b4f78 (“Simplify QA commands”) (but it was still imported for a
  while, until we extended pylint to run over the QA code as well)
- qlang._IsHostname, added in commit f8638e28 (“Detect globbing
  patterns as query arguments”) but never used
- cmdlib._BuildNetworkHookEnvByObject, last usage removed in commit
  f22433c0 (“Locking related fixes for networks”)
- NLD constants, which are obsolete—the nbma project is no longer
  maintained and it's not working with current ganeti

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

This patch is enabling split queries for jobs for gnt-node and rapi
access (only for job listing, not job waiting).

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

This shouldn't live in the main Query module, since it's too node
specific.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

This adds support for job queries, including (basic) unit-tests.

I've tested this for memory and cpu usage as follows:

- 3600 jobs (live queue):
  - via masterd, default: ~1.1s (masterd: ~60MB ram)
  - via confd,   default: ~1.1s (hconfd:  ~25MB ram)
  - via masterd, id only: ~1.0s (masterd: ~57MB ram)
  - via confd,   id only: ~0.2s (hconfd:  ~15MB ram)

- all jobs (128K in total, around 570MB size on disk):
  - via masterd, default: 1m22s (masterd cpu: 48s), masterd: 1.4G ram
  - via confd,   default: 1m16s (hconfd  cpu: 51s), hconfd:  570MB ram peak
    (peak only right before starting luxi send, hconfd decreases in RSS as
    results are streamed over luxi, back to ~18MB after the send)
  - via masterd, id only:  ~49s (masterd cpu: 45s), masterd: 1.3G ram
  - via confd,   id only:  ~39s (hconfd cpu: 35s),  hconfd:  110MB ram peak
    (right before luxi send, decreasing as results are sent, back to ~14MB
    after the send)

Given this, and that in production it's not likely to have hundreds of
thousand of job files, I believe the implementation is safe from this
point of view.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

This is required for QueryJobs implementation; we change
makeSimpleFilter to support both string and integers as names.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

This patch adds implementation for a read-only job queue module,
together with "full" test (as full as can be in a lazy language…).

One note about the behaviour of the job queue is the handling of
opcodes that fail validation: the 'input' opcode actually is a
meta-type, which can hold either a real opcode or a plain JSValue, so
that we can still load jobs with invalid opcodes for querying. The
only downside of this is that, as opposed to Python code, we can't
show the correct summary for such an opcode - we try to parse the
OP_ID but not the extended OP_DSC_FIELD-equivalent.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

I am very very unhappy with this patch. We have to do this, in order
to be compatible with the Python code and behaviour, but there's lots
of duplication; both the Python and the Haskell code will need to be
cleaned up and simplified (we don't need QFF_* at all).

For now though, this implements QffTimestamp and associated code.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Unit tests updated and expanded with an inheritance check.

The flag has no effect yet.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

PVs don't need to be marked as unallocatable, as the code will ignore
partially used PVs. exclusive_storage made a node parameter.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Forgot to fix that after the review of patch
"Haskell/python compatibility test for networks".

Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Since this is a type that behaves non-standard, let's add some more
test for "wrong" input values.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

As OpParams definitions are used at Template Haskell type, they don't
get any coverage (although the values defined by them are actually
used in unittests). However, we can at least test some of the
functions defined in the module.

This patch tests the failure modes of a few of the custom types in
this module.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

"Use comparing" was not needed, and "Use on" can be solved by actually
using on in the single case we have (the patch also reorders imports
in that file).

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

These test the newly-changed behaviour, and fix an inconsistency in
the hs-check target (versus hs-coverage).

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

Currently, this doesn't abort if multiple arguments are passed, and
the error message for errors during file read has a duplicate "Error".

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

This just ties in all the pieces so far and enables bash completion
for it.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

This patch adds support for parsing the command list out of a binary
(very strict format), and then iterating over that and building the
sub-commands options/arguments.

It also does a bit of general cleanup in the script.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

This will allow build-bash-completion to understand what is the list
of commands that each binary supports, and then provide recursively
completion support for each command.

Note: by "fake", I mean we use hand-built output and hardcoded
"--help-completion" matching, as opposed to automatically doing it via
regular options. It could be possible in the future to use a
two-staged command line parser, but for now not sure if it's worth it.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

No need to do a boolean-based lookup, we can use the regular
association list lookup.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

This allows usage information to display nicer help (like in Python).

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

Currently, the mon-collector binary is the only multi-command binary
we have in Haskell. Not surprisingly therefore, its command line
handling is not as robust as we need, not being able to support
standard "--help" commands.

This patch introduces generic multi-command handling, similar to the
Python functionality (some parts missing, will be added in future
patches), and switches mon-collector over to it.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

The signature of the personality definitions is so ugly that, at
least, we should hide it a bit behind a type alias.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

Currently, the mon-collector binary uses the HTools/CLI module, which
is OK but mean it links in lots of htools code. By copying that module
to DataCollectors/CLI and removing the unneeded code, we reduce the
number of modules it depends on fro 20 to 12, meaning both a shorter
compiler time (24s to 9s) and a smaller binary (~9.6MB to ~7MB).

Also fixes a typo in the original HTools/CLI module, thanks Michele!

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

As noted by Guido, there are problems when using the hs-prof and
hs-prof-quick targets in the default configuration (compiling all
programs). The errors manifest in the form of wrong symbols during
compilation.

I knew that the hs-prof targets, which compiled multiple objects with
the '.o' suffix, could be problematic; but the objects that are
actually needed in the Template Haskell phase are very standard and
don't differ between the binaries (Constants, BasicTypes, JSON,
THH). What I didn't realise (although it's obvious) is that also the
hs_prof_quick targets (the final binaries) are also compiled with a
single suffix ('.prof_o'), which means that the object files are
actually compiled for the last binary.

This means that targets later in the HS_ALL_PROGS list would work
correct, but early targets, especially htools/htools, would fail.

So the obvious, and the single simple solution is to make these two
rules only work on a single binary at a time. This should be fine,
since one is looking at a specific problem usually, and it has the
advantage that the hs-prof step is much faster (since it wasn't
buildable in parallel anyway).

Thanks to Guido for finding and making the initial diagnose on this!

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

If an instance is offline we definitely shouldn't start it up.
But shutting it down, should it be up by mistake is not "that" bad.
Still, we only allow it with --force, as it still performs an action on
an instance we shouldn't touch. This should make everybody happy.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

If an instance is running (eg. ERROR_up) and at the same time offline,
there's no way to either shut it down or reonline it. This allows
onlining it. Offlining is still disabled for running instances.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Some fixes, added more information in a few points, removed a stale (5+
year old) TODO comment.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

This can be more useful than the "make" version.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Commit 141d1489 was a bit too enthusiastic. The three parameters added to
the list of parameters to be checked default to a value not evaluating
to false, leading to a failure on cluster initialization.

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

LEVEL_NODE_ALLOC should be aquired too if LEVEL_NODE is ALL_SET.

Signed-off-by: Dimitris Aragiorgis <dimara@grnet.gr>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

This patch contains the HUnit test that checks the compatibility
of Haskell-generated networks with the python code. For that the
generation of test instances of networks was enhanced to meet
the validation steps of the python implementation. Also, so far
networks were generated at two different places in the code; this
is now consolidated.

Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

“_SPICE_ADDITIONAL_PARAMS” is supposed to be the full list of
SPICE-related KVM hypervisor parameters with the exception of
“HV_KVM_SPICE_BIND”. The new test checks if all parameters starting with
“HV_KVM_SPICE_*” are included. Three previously missing parameters are
added to the list.

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

This list is static at runtime and doesn't need to be recreated every
time.

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

When non-DRBD disks are used for an instance,
“lu.needed_locks[locking.LEVEL_NODE]” is set to “locking.ALL_SET” (which
is None). The assertion will then fail as None evaluates to False.

Reported by Constantinos Venetsanopoulos.

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Doing so avoids triggering an assertion in mcpu which cross-checks the
node lock and node allocation lock acquisition mode.

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Commit f0f8d060 (“Show old primary/secondary node on disk
replacement”) added two wrong uses of feedback_fn, which results in
log entries such like these:

      "log": [
        [
          7, [1351258326, 466214], "message", "Replacing disk(s) 0 for instance 'instance1.example.com'"
        ],
        [
          8, [1351258326, 749391], "Current primary node: %s", "node1"
        ],
        [
          9, [1351258326, 774829], "Current seconary node: %s", "node4"
        ],
        [
          10, [1351258329, 620410], "message", "STEP 1/6 Check device existence"
        ],

(the first and fourth are OK, the 2nd and 3rd are broken).

The API of jqueue._OpExecCallbacks.Feedback should probably be changed
to prevent such an occurrence in the future.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

Some confd types were not tested for serialisation, so let's move
these (type-related) tests to their own module and extend them.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Commit e821050d (“Switch the Luxi interface from Strings to
ByteStrings”) was designed to optimise the receive interface, but has
an unfortunate side-effect: when sending non-trivial messages, it
means that both the entire String and the ByteString versions must be
in memory at the same time, leading to much increased memory usage.

By changing the "hPut" from strict to lazy ByteStrings, it means that
both the String and the ByteString values can be evaluated lazily,
with significant effects: for a test query answer, instead of having
a peak from ~600MB to 1.4G during the entire Luxi send operation,
memory consumption actually decreased during the send operation, as
the ByteString chunks are released individually and even the backing
storage of the items that create the JSON string serialisation is
released lazily as well. So instead of slow growth 10→550MB then quick
peak to 1.4GB during Luxi send, we now have an even slower growth to
~580MB and then decrease of memory as the Luxi send progresses.

The only downside is of a small increase in CPU time of a few percents
for the above case; for our use cases, I think this is much desirable.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

This allows us to ensure that query results are strict as we build
them, instead of being lazy and only evaluated when the reply is sent
over the Luxi interface.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>