Since hlint doesn't check style issues but rather code only issues,
let's add a simple Makefile check for too-long lines, and fix the
single bad case we have currently.

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

Indentation within rules is fixed to always be 2 spaces (as opposed to
mixed tabs/2 spaces/4 spaces). Additionally, the check-dirs rule,
which was more complex than necessary, has been simplified: moved
auto4mte.cache to DIRCHECK_EXCLUDE, and simplify the find call and the
exit condition (no need for if test …; then exit 0; else exit 1; fi).

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

The current uniqueness algorithm (generate random node names, suffix
them with node index) is actually wrong: a node named "21" at index 5
will end up with the same name as a node named "2" at position 15.

To fix this, we also add a character from a different "set" ("-"), so
that such mixups can't happen again, and also add an explicit check
for it.

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

All resources are acquired and opportunistic instance creations would
fail. Also add a TODO.

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

Opportunistic locks are not yet used. This patch changes
LUInstanceCreate to acquire the node allocation lock to avoid conflicts
with other opcodes acquiring many node locks.

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

See code for an explanatory comment. The lock is released as soon as
possible.

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

LUNetworkConnect/Disconnect looks up a nodegroup's instances for
conflicting IPs. To do so, use GetNodeGroupInstances() and
GetMulitInstanceInfo().

Additionally, check if the correct locks were acquired.

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

The network management code needs to verify a MAC address prefix.
Instead of (ab)using NormalizeAndValidateMac, clean code should be used.
Unit tests for NormalizeAndValidateMac are updated and new ones for
NormalizeAndValidateThreeOctetMacPrefix are added.

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

Instead of having two different lists of directories, now there is only one.

Signed-off-by: Michele Tartara <mtartara@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

This code is going to be used by a new utility for setting up the node
daemon. Unit tests are updated/added.

Additionally, the certificate and key stored in “server.pem” are
verified, too.

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

Similar to “share_locks”, a new dictionary containing booleans for each
locking level is added to “cmdlib.LogicalUnit”. Logical units wanting to
make use of opportunistic locks will be able to configure this
dictionary accordingly.

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

Just forwarding the parameter, nothing more.

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

This patch adds a new parameter to “LockSet.acquire” named
“opportunistic”. When enabled the lockset will try to acquire as many
locks as possible, but it won't wait for them (with the exception of the
lockset-internal lock in case the whole set is acquired). This is
implemented by using a timeout of 0 seconds when acquiring individual
locks. Commit 03c5291c made such acquisitions significantly cheaper.

The most complicated code included in this patch is probably the helper
function used to determine which mode to use and which timeout functions
are needed.

Full unit tests are included for the new and changed code
(“locking.py”'s overall coverage is at 97%).

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

Configuring a node daemon on a newly added node will need all ssconf
values.

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

Commit 224ff0f7 modified the node SSH setup to use the system's SSH
client. Before that Paramiko was used. It's not entirely clear whehter
the latter ever supported IPv6 properly, but with this patch
“bootstrap.RunNodeSetupCmd” is changed to use it if configured. The code
for doing so is taken from “bootstrap.SetupNodeDaemon”, which is going
to use “bootstrap.RunNodeSetupCmd”.

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

Part of the code used for running “prepare-node-join” can be re-used for
running a tool to configure the node daemon.

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

A new utility for configuring the node daemon will support a dry-run
mode. This patch adds the necessary functionality to
“ssconf.SimpleStore” and provides comprehensive tests for
“SimpleStore.WriteFiles”. To enable the latter, a testing-only parameter
specifying the lockfile was added to the class.

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

The new utility for configuring the node daemon will have to check
whether it received valid ssconf names.

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

Add the design document for opportunistic locking.

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

This function can be applied to any type of class "Element" but its
internal variables used to refer to a list of tuples containing these
elements and their names as "nodes". Moreover it built a list of name to
indices tuples referring to it as "na" (which is not very clear).

Thus nodes becomes name_element, and na name_idx.

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

In case conflicts are checked, log warnings if nodes' IPs cannot
be reserved.

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

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

Ensure that locks are held only if needed.

Add conflicts_check in OpNetworkAdd. This is needed if we want to
check whether nodes/master IPs are included in network.

Depending on conflicts_check value, we have to hold node/instance locks
during OpNetworkAdd/OpNetworkConnect or not.

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

The serialization itself is done by Text.JSON, so the tests deal with checking
that Text.JSON objects are created correctly from the DRBD parser data
structures.

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

The serialized JSON is not a 1:1 dump of the data structures
populated by the parser. This is done intentionally, with the
aim of producing a more stable and more meaningful output to
be used by the (future) monitoring agent and stand-alone data
collectors.

Also:
* shorten the names of some data types.
* change data type for the fields of the Time data structure

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

The new support functions in module Ganeti.JSON will be used by the next
commits to build JSON objects with optional (Maybe) fields, excluding
them when they contain Nothing.

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

This improves Issue 325 - new runtime and memory consumption is about
1/10 compared to before.

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

Given a node list in input, we get an instance that had nodes in it.

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

This takes an instance generator and produces a possibly empty list of
instances.

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

This generates a minimum of one node, because legal clusters never have
zero nodes.

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

hbal doesn't have any "-o" option, but on the other hand supports -S
(described below, but not present in the short summary)

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

This is used only once when testing Cluster.hs, but having it abstracted
clarifies there what that call is about, makes that test shorter, and
allows us to better do refactoring of the main genInstanceSmallerThan
generator.

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

The new one is more consistent with the rest of the nodegroup
terminology and more generic for use in different htools.

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

It turns out that in newer GHC versions, the suffix shouldn't be
prefixed with a dot; otherwise name mangling is broken.

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

Currently, the And and Or filters use very nice code, e.g. in case of
OrFilter:

   any id <$> mapM evaluateFilter flts

However, looking at the memory profiles shows that application of
any/id to monadic values via '<$>' does not work nicely, losing the
'early' success property. This results in too much memory being used
for thunks in monadic sequencing.

Rather than trying to add more strictness (not sure exactly how, TBH),
switching these to explicit recursion solves the problem, since we
take the 'early' exit problem in our hands and we are explicit about
it.

Memory usage in case of big (e.g. 1000 'Or' elements) is reduced
significantly, and thus also the runtime.

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

Currently, the query operators (binop, etc.) create thunks, instead of
forcing the evaluation of the simple boolean results. This results in
higher than needed memory use.

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

Since this program is not built normally, a few recent commits broke
it:

- 707cd3d7 (“Use exitErr instead of explicit error message and
  exitWith”) introduced wrong syntax (missing '$')
- 29a30533 (“Remove unsafePerformIO usage”) changed the signature of
  G.Path.clusterConfFile
- 5183e8be (“Convert query path from string errors to GanetiException”)
  changed the signature of 'getNode'

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

I've been bitten a couple of times with arbitrary opcodes working on
UTF-8 locale, but failing on buildbot (ASCII). So let's add an
explicit test that checks always (even with UTF-8) for correct
arbitrary values, showing explicitly which opcodes fail.

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

It turns out that optimising 'read' derived instances (via -O) for
complex data types (like OpCode, or the various objects) can be slow
to very slow. Disabling such instances results in (time make
$all_our_haskell_binaries) large compile-time savings and also smaller
(unstripped) binaries (by a significant amount):

ghc 6.12:        time  htools sz  hconfd sz
  with read:    4m50s 12,244,694 14,927,928
  no read:      3m30s 10,234,305 12,536,745
ghc 7.6:
  with read:   14m45s 13,694,761 15,741,755
  no read:      3m40s 11,631,373 13,245,134

So let's remove these instances, since we never use read in production
for our custom types, and even when debugging in GHCI, we can simply
use the 'show' representation to create the types, without needing to
actually parse from strings.

Note: for the very slow ghc 7.6 compilation time, I filled a ticket
(ghc #7450), since it is surprising(ly slow).

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

This is the last inline-written field; I've done it separately from
the last patch due to changes being needed in HTools (field changed
type).

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