I have just noticed a bug in the Telecom Log Service IDL. The
"Log" interface defined in the "DsLogAdmin" module defines "set_qos"
and "get_qos" operations. And, the DsNotifyLogAdmin::NotifyLog
interface inherits both the DsLogAdmin::Log interface (indirectly,
through inheritence of DsEventLogAdmin::EventLog) and the
CosNotifyChannelAdmin::EventChannel interface. Note that the
CosNotifyChannelAdmin::EventChannel interface inherits from
CosNotification::QoSAdmin, which also defines "set_qos" and "get_qos"
operations! This is invalid IDL inheritence as of CORBA 2.3!

The same problem exists for the DsTypedNotifyLogAdmin::TypedNotifyLog
interface. It indirectly inherits both DsLogAdmin::Log and
CosNotification::QoSAdmin, and thus two versions of "set_qos" and
"get_qos".

There are a couple of possible solutions to this problem, but perhaps
the easiest would be to simply rename the "set_qos" and "get_qos" operations
defined in the Log interface as "set_log_qos" and "get_log_qos",
respectively. If only there were an RTF that could do this...

see above

The specification says that "The write_recordlist() operation is provided ass a convienience operation to allow the output of a query of one log to be written to another log. In this case, only the "info" field (event content) of the RecordData struct is written to the log, the LogRecord id and logging time will be assigned by the log.". This is the only place in the spec that RecordData is used, I believe that LogRecord was intended. Also, it is unclear whether the attr_list field should be copied. While it says "only the info field" is written, attr_list is among the fields listed that will be "assigned by the log". For what it's worth, the Orbix Log Service implementation appears to copy the attr_list. Their Enterprise Messaging Guide for ORBIX 6.2 says "write_recordlist() is functionally identical to write_records(). It writes data directly to the log and raises the same exceptions. The major difference is that the record's data is stored in a LogRecord. This allows you to add a series of name/value pair attributes to assist in querying the log". The TAO log service implementation, which I currently maintain, also copies the value of attr_list.

The specification says that "The write_recordlist() operation is provided ass a convienience operation to allow the output of a query of one log to be written to another log. In this case, only the "info" field (event content) of the RecordData struct is written to the log, the LogRecord id and logging time will be assigned by the log.". This is the only place in the spec that RecordData is used, I believe that LogRecord was intended. Also, it is unclear whether the attr_list field should be copied. While it says "only the info field" is written, attr_list is among the fields listed that will be "assigned by the log". For what it's worth, the Orbix Log Service implementation appears to copy the attr_list. Their Enterprise Messaging Guide for ORBIX 6.2 says "write_recordlist() is functionally identical to write_records(). It writes data directly to the log and raises the same exceptions. The major difference is that the record's data is stored in a LogRecord. This allows you to add a series of name/value pair attributes to assist in querying the log". The TAO log service implementation, which I currently maintain, also copies the value of attr_list.

The specification states "If the capacity of a log exeeds that of one of its log capacity alarm thresholds, ..." This wording is confusing: The log capacity (as set by set_max_size()) is not compared with the capacity alarm thresholds; it does not state that the ThresholdAlarm is sent once when the threshold has been met, which may lead some to believe the alarm events will be sent continuously; and the word "Exceed" means that a alarm would never be sent for a threshold of 100%. I think that something like "If after a log record has been written, the current size of a log, as expressed as a percentage of max log size, is greater or equal to the next log capacity threshold, then a ThresholdAlarm event is generated ..." might be clearer.

I am in the process of updating TAO's Logging Service implementation. Section 1.2.5.1 stats "The Telecom Log Service proposes to define the following OMG Standard exception codes for use with the CosEvent and CosNotification push operations..." and goes on to mention a LOGFULL and LOGOFFDUTY minor code for the NO_RESOURCE [sic] SystemException, a LOGLOCKED minor code for the NO_PERMISSIONS SystemException, and a LOGDISABLED minor code for the TRANSIENT SystemException. TAO does not define these minor codes, so I checked the latest CORBA spec (3.0.3 / formal/04-03-12) and they are not in Appendix A either. What exactly does the spec mean "proposes to define" and what should an implementer do to handle these errors in the here and now.

Section 1.4.4 states that the "AttributeValueChange event is generated by a log factor when one of its log changes one of changes one of the following attributes..." The description of the set_* methods state "An AttributeValueChange is generated whenever the * is set". The former implies that the event will be sent if the value is changed, the latter imples the event will be sent even if the value is the same.

· Chapter 2.5, page 2-33. The paragraph of the second bullet point can also contain the TypedRecordIterator.

I am the current maintainer of the TAO logging service implementation. The specification is unclear what the interactions between set_max_size(), set_log_full_action(), and set_capacity_alarm_thresholds() have on the current processing of capacity alarms. For example, consider a log with DsLogAdmin::halt log full action and 0 (ie. infinate) max size, and a capacity alarm threshold list of [50, 75, 90, 100]. If the log has 80KB of records, and the max size is set to 100KB, are ThresholdAlarms supposed to be sent for crossing 50 and 75%, or should a threshold alarm only be sent once the log grows to 90%? Similar questions arise when the log full action is changed to/from DsLogAdmin::halt from/to DsLogAdmin::wrap, as the capacity alarm threshold processing changes from a percentage to a "gauge". Or when the capacity alarm threshold list is changed from one set of values to another.

I'm working on TAO's Log Service implementation, and I'm not sure how I'm supposed to support QoS properties. The spec for get_log_qos() method states it "returns a list of the quality of service properties supported by the log." This is different from the other get_* methods, which return the current value of the settings. If this is the intent, it should be emphasized, and perhaps a new method introduced to fetch the current QoS settings. The three defined QoS settings QoSNone, QoSFlush, and QoSReliability are mutually exclusive. But unlike the Notification Service Spec's section on QoS settings, there is no guideance wrt. semantics of a property list with multiple properties or what should be returned in the denied list in the UnsupportedQoS exception.

· Chapter 2.3.1, page 2-26. It is more correct if it is said that BasicLogFactory creates BasicLog but not Log (similar to that EventLogFactory creates EventLog).

· Chapter 2.2.7.2, page 2-25. destroy operation is not only inherited from CosEventChannelAdmin::EventChannel but also defined in BasicLog interface. It is worth to mention it, since destroy operation can also be invoked on BasicLog objects.

· Chapter 2.4.1, page 2-29. Definition of the ObjectCreation struct contains logref field. This is different from the same definition in appendix 1 on page A-7. I think the correct version is on page 2-29. Then ObjectDeletion struct (page 2-30) does not have logref field. Is it OK? If yes, then "NOTE" in appendix 1 on page A-7 is wrong and not needed at all. If not, then the definition of that struct both on page 2-30 and in the appendix 1 should be modified.

The IDL for the ThresholdAlarm event defines three constants for the percieved_severity field: critical, minor, and cleared. The text says the "percieved_severity is minor if log is not full, and critical otherwise". If these are the only valid values, why is there a definition for cleared?

· Chapter 2.4. Within the whole chapter in the description of every generated event "the log field" is used instead of "the logref field". Should be changed in every occurrence.

· Chapter 2.2.4.4, page 2-11. To have one style in the document and make it more readable I propose to change ulong to unsigned long.

3. Appendix 1, page A-5 has missing definition of write_recordlist method in the Log interface and has doubled definition of write_records method.

All these issues are applied to version 1.0 as well (formal/2000-01-04).

· Chapter 2.2.7.1, page 2-24. What is so resource consuming in the copy and copy_with_id operations? Why it can raise NoResources exception and any factory cannot? Factories also create logs and the log server also can have not enough resources to do it, but according to the specification factories do not raise NoResources exception. If NoResources should leave there then appendix 1 should be corrected since definitions of these copy operations do not contain NoResources exception in the list of raised exceptions and appendix 1 does not have definition of NoResources exception at all (should be put in DsLogAdmin module). Or if NoResources can be omitted then description of those copy operations should be modified

· Chapter 2.2.6, page 2-24. It is not clear for reader what is DAG topology. I propose to explain this. At least write what it stands for, like "DAG (directed acyclic graph)".

· Chapter 2.2.5.5, page 2-22. set_record_attribute operation does not raise InvalidGrammar and InvalidConstraint exceptions. These exceptions should be deleted from the description of the operation

2. Page 2-19. Description of write_recordlist says "In this case, only the "info" field (event content) of the RecordData struct is written to the log, the LogRecord id and logging time will be assigned by the log." Why doesn't it say anything about attr_list field of RecordData struct? Seems to me that this field (list of name/value pairs) can also be modified and written to the log.

1. Page 2-19 has missing LogOffDuty exception in the signatures of write_records and write_recordlist methods.

Currently there is:
void write_records(in Anys records)
raises(LogFull, LogLocked, LogDisabled);
void write_recordlist(in RecordList list)
raises(LogFull, LogLocked, LogDisabled);

Should be:
void write_records(in Anys records)
raises(LogFull, LogOffDuty, LogLocked, LogDisabled);
void write_recordlist(in RecordList list)
raises(LogFull, LogOffDuty, LogLocked, LogDisabled);

· Chapter 2.2.5.4, page 2-21. delete_records operation does not raise InvalidAttribute exception. This exception should be deleted from the description of the operation.

· Chapter 2.2.5.1, page 2-19. What are the values for minor codes: LOGFULL, LOGOFFDUTY, LOGLOCKED, and LOGDISABLED? Should those be defined in IDL?

The text for the copy() and copy_with_id() operations defined on the DsLogAdmin::Log
interface should specify that a DsLogNotification::ObjectCreation notification will
be generated as a result of the copy operation.

Summary: The specification text is unclear and should make a note in
section "3.4 Log Generated Events" to explain that the manner
in which logs send log generated events to their corresponding
log factories is an implementation detail.

The idea is that applications (log/LogFactory clients) are only consumers
of log emitted events, they do not send events to the factory. Therefore, the
LogFactory does not need to expose the SupplierAdmin interface. How the log
actually forwards events to the factory is left up to the implementation -
use a SupplierAdmin object, or other means.

resolved, see revised text below

Summary: In the specification text, the BasicLogFactory::create() and
BasicLogFactory::create_with_id() operations have an extra
argument, in CapacityAlarmThresholdList thresholds, which does
not appear in the specification IDL. The reference to thresholds
in the specification text should be removed from the definitions
of BasicLogFactory::create() and BasicLogFactory::create_with_id().

The paragraphs in the specification text that describe the create()
and create_with_id() operations for the other log factories
(ie. EventLogFactory) should explain that they have an additional
argument, thresholds, besides the ones defined in the BasicLogFactory.

resolved, see revised text

Summary: In the specification text, all references to "LogFactory" should be
changed to "BasicLogFactory" to match what is defined in
the specification IDL.

completed, see revised text

Summary: there are a few enums defined in the
IDL. Perhaps it would be a good idea to adopt the
current OMG style of using CONST declarations for
these values as this allows easy extension of the values (which
enums do not). It is also nice to define the const types
as signed and to reserve all positive values for OMG
use and leave the negative values for vendors.

The following are the enums defined in the Specification IDL and proposed changes:

There are various Typo's that need to be corrected:
Page 1-5 "Allows a log factory emits"
Page 1-5 "Allows multiple push or pull consumers can register"
Page 2-20 "An InvalidAttribute exception..." (delete_records does not raise Invalid exception)
Page 2-21 "interface Iterater"
Page 2-21 "When an invoking the get() operation..."
Page 2-22 "...initialized to the save values as the log..."
Page 2-26 "...TypedNotifyLogFactory interfaces create() new EventLog..." (no parens)

Correct typos.

Summary: The specification text is unclear to a log"s behavior when off_duty.
What should the log do when it"s not functioning due to being
outside the interval or shut down according to the weekmask? I assume it"s
still available for adjusting weekmask parameters, but does it just drop
log requests on the floor?

rersolved, see revised text below

Summary: The specification text is unclear to a log"s behavior when disabled.
If the log is disabled ("due to some run time catastrophy"), how should
the log respond to requests? Since it is active enough to respond to
"get_operational_state ()" requests, shouldn"t it throw exceptions? Is it
allowed to drop log requests on the floor in that case? It could throw a
LogLocked exception, but in that case, should the log lock when it enters
the disabled state?

resolved, see revised text

The specification does not define a StateChange or a ProcessingErrorAlarm event
which are referenced in X.735 section 11.2.3, "Notifications".

resolved, se revised text below

The specification is inconsisent in specifying which attributes generate
AttributeValueChange events. Currently AttributeValueChange events are
generated for:
capacityAlarmThreshold
logFullAction
maxLogSize
startTime
stopTime
weekMask
filter

An AttributeValueChange event should also be generated when the other
attributes of a log are changed:
administrativeState
maxRecordLife
qualityOfService
forwardingState

resolved, see below

Summary: Documents: telecom/98-12-04 (text)
telecom/99-01-05 (IDL)

In module DsLogNotification, the event to be sent on log deletion is
defined as

struct ObjectDeletion

;

Here, "logref" is a reference to the Log object that has just been deleted,
and hence cannot be used for anything – would it be better to abolish this
field? (Even though doing so would reduce the symmetry with ObjectCreation).

Remove logref from the ObjectDeletion struct.

I think the OMG Telecom Log Service IDL (DsLogAdmin and DsNotifyLogAdmin
modules) isn't in accordance with the CORBA 2.6 "OMG IDL Syntax and
Semantics" specification.

Indeed the DsNotifyLogAdmin::NotifyLog interface inherits from both
DsLogAdmin::Log.get_qos and CosNotification::QoSAdmin.get_qos
operations, this is illegal according to section 3.7.5 (page 3-20) of
the CORBA 2.6 specification.

I've tried to compile the OMG Telecom Log Service with 2 differents ORBs
(ORBacus 4.0.5 and JacORB 1.4beta2) without success.

There is a typo on the first sentence on page 2-13. The words "enabled lock" should be "enabled log" in the following sentence : The stop field indicates the date and time at which an unlocked and enabled lock stops functioning.

Summary: There seems to be a slight conflict on pg 3-16: "New events delivered to a
halt log will simply be discarded, these events may still be forwarded if
the log forwarding state (described below) is on". How can events be
discarded and forwarded?

A log in the halt state only effect the logs ability to log new records.

Summary: Should the Log interface define the get_capacity_thresholdlist()
and set_capacity_thresholdlist() operations? The BasicLog interface
inherits from Log and yet thresholds are meaningless in the context
of the BasicLog Why are they defined in Log and not in event and
notify log? And should BasicLog even bother checking the parameters or
doing anything?

resolved , see text below

A clarification needs to be made when untyped events are inserted and retrieved
from a typed log.

A client can use the standard "push" operation with a TypedProxyPushConsumer as
well as the type event operations defined on the interface generating the event.
What happens if a client invokes the typed_query() operation where some of the events
that pass the constraint were "push"ed as untyped events? Are those records ignored?
Are the type related fields of the TypedEventRecord left empty? Should we raise an exception?

No action to be taken.

Summary: The description of the "in Anys records" parameter of the write_records()
operation on page 3-18 is unclear as to what the records should contain.

The specification text should state that each Any contains the event to be logged.

The Notification Service defines:

module CosNotification {
// ...
interface QoSAdmin

;
// ...
};

module CosNotifyChannelAdmin {
// ...
interface EventChannel :
CosNotification::QoSAdmin,
CosNotification::AdminPropertiesAdmin,
CosEventChannelAdmin::EventChannel

;
};

The net result is that DsNotifyLogAdmin::NotifyLog inherits get_qos()
and set_qos() twice, once from QoSNotification::QoSAdmin, and once from
DSLogAdmin::Log, which is illegal. This makes the Logging Service
unimplementable right now.

Looks like the operations in DSLogAdmin::Log will have to be renamed or
the inheritance structure will have to change.

Assuming a log is persistent, how does one access the logged data after the system has been shut down? It seems important to be able to view this data after the system shuts down to determine if there were any glitches while the system ran or to troubleshoot any problems that occured.

The Log Service spec doesn't state how a client gets access to the
intial references for the service (namely, the various factories).

This issue concerns the Telecom Log Service as described in
the dtc/99-07-02 document.

I think the following statements in the spec are not compatible
together:

"Log forwarding is determined solely by forwarding state, it is not
affected by the log's administrative state, availability state, log
duration, scheduling, or log full action."

"A log may be not be able to create any new log records because it is
full, locked, or disabled. If a log cannot create new log records when
a push operation is invoked, then the push operation should fail and a
SystemException should be raised."

If the log records are forwarded, then throwing system exceptions to
the supplier doesn't make sense.

The AB raised an issue regarding an ambiguity on the proper value
for the availability status, when the log is in the wrap condition.
The FTF editor added the following sentence at the top of page 2-12
of to telecom/99-05-01, in an effort to clarify the AB issue:

"When a log is full its availability status will change to
indicate the log full condition. If a log's full action is set to
'wrap', it will never be in 'full' condition."

This same ambiguity was discussed the ITU-T group which
originally wrote the X.735 spec, and their proposed resolution
has the opposite affect:

Add the following explanation to the availability status value as a
separate para after the text discussing wrap event behaviour :

"The availability status indicates "logfull" when the capacity of
the log is exceeded. In the case where halt behaviour is
specified for the log the value implies no more records can be
added. If wrap behaviour is specified, the value continues to
remain as log full even though forthcoming records will be added
by overwriting existing records. The wrap mode combined with
logfull condition indicates that old records are lost as
overwriting takes place. If records are deleted from the log, the
logfull status value is removed from the availability status in
either halt or wrap mode."

The reason the ITU group has recommended "full" status is that it
gives the information that records are being overwritten. When
the log is full, records will be lost in response to forthcoming
events. In Halt mode new records are not logged (i.e., lost),
while in the wrap mode the oldest records are overwritten (i.e.,
lost). If a client comes and deletes records, then
availabilitystatus can be reset. Otherwise one needs to have a
different behaviour for halt and wrap in terms of the value for
availability status.

Accept the ITU-T proposal, when in wrap mode, the log is considered to be full.

would like to formally raise the following Issue for resolution
by the FTF (jointly with the ITU-T group responsible for X.735)

Location: section 2 : Log full action, page 2-12

The AB raised an issue regarding an ambiguity on the proper value for the
availability status. When the log is in the wrap condition. The FTF input text
added the following sentence:

This is a duplicate of Issue 2753 – closed

Summary: This behaviour does not seem correct. Under a wrap condition, a
capacity alarm gauge threshold should retrigger once in each
period between wraps, but if the counter resets every time it
crosses the thresold it will trigger more often. Thus the reset
of the gauge , under wrap conditions, needs to be reworded as
follows:

"
When a log object is created with the wrap option the capacity
threshold alarms are triggered as if coupled to a gauge that
counts from zero to the highest capacity of the log defined
(reached each time the log wraps) and then resets to zero.

If a log wraps, and no records are deleted to free space,
capacity threshold events shall be emitted only when all the
records in the log prior to the previous wrap occurance are
completely overwritten. In other words, the gauge threshold is
reset to zero each time the end of log marker is crossed by the
addition of a new log record. For example if a log has a capacity
of 2 Gigabytes a wrap event will occur only when every time 2
Gigabytes of data have been written to the log.
"

resolved, see revised text below