Tuesday, November 17, 2026

Oracle’s November Patch Storm: What EBS, SOA and WebLogic Teams Must Fix First

 

Evidence Pack


Share of new patches that are remotely exploitable without authentication

Oracle APEX
2 / 2
Java SE
11 / 11
Fusion Middleware
53 / 71
REST Data Services
5 / 8
Database Server
3 / 7
E-Business Suite
19 / 71

Source: counts stated in each risk-matrix introduction. EBS is the lowest share, but its 19 are in customer-facing modules.

EBS: 71 new patches by CVSS band

Critical 9.0+
6
High 7.0–8.9
45
Medium 4.0–6.9
20

Counted from the EBS risk matrix. No Low (below 4.0) entries. Module spread: Mobile Field Service 11, Purchasing 9, iProcurement 7, Sales Online 6.

Attack chain and where November’s patches sit

Edge / web tierOHS · APEX · ORDSEBS app tieriStore · AOL · OAF · ISGMiddlewareWLS · OSB · OAM · MFTMgmt planeOEM Agent Next GenDatabaseCreate Session path9545 · 1148499.8 unauth114880 · 114924114925 · 114863 (9.8)101691 (9.3) · 114876 (8.6)91569 · 114824114835 (10.0)91470 (9.8, T3/IIOP)10050 (9.1)unauth, HTTP114819 · 114848114822 (9.9)All IDs are CVE-2026-… Treat this column group as one release.

The week in order

  1. Oct 7–13 Fireworks AI vendor breach; Uber told Oct 9, confirms data access Oct 13
  2. Nov 2 cPGuard update channel hijacked; GitBook credentials accessed
  3. Nov 10 Barracuda ESG chain and Chrome flaw added to CISA KEV; Uber files notice
  4. Nov 13 Uber breach becomes public; UnrealIRCd flaw added to KEV
  5. Nov 16 Optimizely script injection reaches 178 organizations
  6. Nov 17 Oracle CSPU: 655 patches
  7. Dec 15 / Jan 19 Next Oracle releases (CSPU / CPU)

CVE evidence

Each row is taken from Oracle’s risk matrices. “Unauth” means remote exploit without authentication = Yes. None of these is flagged by Oracle as exploited, and I found no public report tying this week’s incidents to them.

CVE-2026-Product · componentCVSSProfile from matrix
114880EBS · Document Mgmt & Collaboration · Attachments9.8Unauth · HTTP · no priv · no UI · C/I/A High
114924EBS · Application Object Library · Core9.8Unauth · HTTP · C/I/A High
114925EBS · Applications Framework · REST Services9.8Unauth · HTTP · C/I/A High
114863EBS · Universal Work Queue · Media Integration9.8Unauth · HTTP · C/I/A High
101691EBS · Mobile Field Service · Admin UI9.3Unauth · HTTP · scope changed · C High, I Low
114876EBS · iStore · Shopping Cart8.6Unauth · HTTP · scope changed · C High
114918EBS · Concurrent Processing · BI Publisher Integration8.7PR High · HTTP · scope changed. Also addresses CVE-2025-61882
114903EBS ISG · Web Service Provider7.4Unauth · SOAP · attack complexity High
114901 / 114904EBS ISG · SOA Provider7.5 / 8.2PR Low, HTTP, AC High / local, PR High, scope changed
114905EBS · Workflow · Notification Mailer5.9Unauth · HTTPS · AC High · C High
114819 / 114848Database · Oracle Text9.9Create Session · Oracle Net · scope changed. Versions 19.3–19.33, 21.3–21.24, 23.4.0–23.26.3 (114819 stops at 19.32)
114822Database · RDBMS9.9Create Session · Oracle Net · 23.4.0–23.26.4
114849APEX 24.1.22–26.2.0 and ORDS 24.2.0–26.3.09.8Unauth · HTTPS · C/I/A High
114850APEX · General9.8Unauth · HTTPS · C/I/A High
9545OHS · WebLogic Proxy Plug-in (curl)7.5Unauth · HTTPS · C High · 12.2.1.4.0, 14.1.2.0.0
91569WebLogic · TopLink Integration10.0Unauth · HTTP · scope changed · 12.2.1.4.0–15.1.1.0.0
91470WebLogic · Core9.8Unauth · T3, IIOP · C/I/A High
114824Platform Security for Java · Thirdparty Jars10.0Unauth · HTTP · scope changed
114835Access Manager · Admin Console10.0Unauth · HTTP · scope changed (two more OAM 10.0s: 91500, 91510)
91561 / 91566Service Bus · Core9.6 / 8.8PR Low · RMI, scope changed / HTTP
8763Managed File Transfer · Runtime (Bouncy Castle)9.1Unauth · HTTPS · 12.2.1.4.0
10050Enterprise Manager · Agent Next Gen (Jetty)9.1Unauth · HTTP · 13.5, 24.1

Queries

Adapt to your release and log formats, and test in non-production first. EBS table and column names can vary by version. Patch numbers come from My Oracle Support note KA923.

3.1 Confirm release and patch status (EBS database)

SELECT release_name FROM fnd_product_groups;

SELECT bug_number, last_update_date
FROM   ad_bugs
WHERE  bug_number IN ('&patch1', '&patch2')
ORDER  BY last_update_date DESC;

3.2 New users and responsibilities in the last 14 days

SELECT user_name, creation_date, last_update_date, last_logon_date
FROM   fnd_user
WHERE  creation_date > SYSDATE-14 OR last_update_date > SYSDATE-14
ORDER  BY last_update_date DESC;

SELECT u.user_name, r.responsibility_name, d.creation_date
FROM   fnd_user_resp_groups_direct d
JOIN   fnd_user u ON u.user_id = d.user_id
JOIN   fnd_responsibility_vl r
       ON r.responsibility_id = d.responsibility_id
      AND r.application_id    = d.responsibility_application_id
WHERE  d.creation_date > SYSDATE-14
ORDER  BY d.creation_date DESC;

3.3 Host-method executables created recently (possible command execution path)

SELECT e.executable_name, e.execution_file_name, e.creation_date, u.user_name AS created_by
FROM   fnd_executables e
JOIN   fnd_user u ON u.user_id = e.created_by
WHERE  e.execution_method_code = 'H'
AND    e.creation_date > SYSDATE-30
ORDER  BY e.creation_date DESC;

3.4 Supplier bank-account changes (account masked)

SELECT '****'||SUBSTR(b.bank_account_num,-4) AS acct, b.last_update_date, u.user_name AS changed_by
FROM   iby_ext_bank_accounts b
LEFT   JOIN fnd_user u ON u.user_id = b.last_updated_by
WHERE  b.last_update_date > SYSDATE-14
ORDER  BY b.last_update_date DESC;

3.5 Unusual attachment uploads (staging or web-shell drops)

SELECT file_id, file_name, upload_date, ROUND(DBMS_LOB.GETLENGTH(file_data)/1048576,1) AS mb
FROM   fnd_lobs
WHERE  upload_date > SYSDATE-14
AND   (LOWER(file_name) LIKE '%.jsp' OR DBMS_LOB.GETLENGTH(file_data) > 52428800)
ORDER  BY mb DESC;

3.6 Database unified audit: failed logons, privilege changes, canary account

SELECT event_timestamp, dbusername, userhost, client_program_name, return_code
FROM   unified_audit_trail
WHERE  action_name = 'LOGON' AND return_code != 0
AND    event_timestamp > SYSTIMESTAMP - INTERVAL '14' DAY
ORDER  BY event_timestamp DESC;

SELECT event_timestamp, dbusername, action_name, object_name, userhost
FROM   unified_audit_trail
WHERE  action_name IN ('CREATE USER','ALTER USER','GRANT','CREATE DATABASE LINK')
AND    event_timestamp > SYSTIMESTAMP - INTERVAL '14' DAY;

-- Honeytoken: a locked decoy account that nobody should ever use
CREATE USER svc_isg_backup IDENTIFIED BY "replace-with-long-random" ACCOUNT LOCK;
SELECT event_timestamp, userhost, client_program_name
FROM   unified_audit_trail
WHERE  dbusername = 'SVC_ISG_BACKUP' AND action_name = 'LOGON';

Confirm the logon-failure audit policy (ORA_LOGON_FAILURES) is enabled before relying on this.

3.7 Web tier: OHS and WebLogic access logs (Apache combined format)

# Sources generating error responses on EBS, ISG and ORDS paths
grep -E '/(OA_HTML|webservices|ords)/' access_log* | awk '$9 ~ /^[45]/ {print $1}' | sort | uniq -c | sort -rn | head -20

# ISG callers: compare against your partner allowlist
grep -E '/webservices/(rest|SOAProvider)/' access_log* | awk '{print $1}' | sort | uniq -c | sort -rn
# Splunk-style: bursts or path scanning
index=web sourcetype=access_combined (uri_path="/OA_HTML/*" OR uri_path="/webservices/*" OR uri_path="/ords/*") status>=400
| bin _time span=5m
| stats count dc(uri_path) AS paths by clientip, _time
| where count>100 OR paths>25

3.8 Browser side: Optimizely-style script risk

# Report-only CSP to learn what your checkout page really loads
Content-Security-Policy-Report-Only: default-src 'self'; script-src 'self' https://cdn.vendor.example; connect-src 'self'; form-action 'self'; report-uri /csp-report

# Daily hash of each third-party script; alert on any difference
for u in $(cat third_party_scripts.txt); do echo "$(curl -s "$u" | sha256sum | cut -d' ' -f1)  $u"; done > hashes_$(date +%F).txt
diff hashes_yesterday.txt hashes_$(date +%F).txt

Oracle Critical Security Patch Update Advisory - November 2026

 

Oracle’s November 2026 CSPU:

Oracle’s November 2026 Critical Security Patch Update (released 17 November) delivers 655 new patches. The count matters less than the timing. This post covers what’s in the patch for E-Business Suite (EBS), the SOA and middleware layer, and the governance work around it. It also covers how this week’s attacks change the order in which you should do things.

  • EBS: 71 patches, 19 unauthenticated-remote. Fusion Middleware: 71 patches, 53 unauthenticated-remote. Database: 7 patches, 3 unauthenticated-remote.
  • Oracle’s advisory doesn’t flag any of these CVEs as exploited. None of the incidents I found this week involve them either. But Oracle estates have been hit repeatedly by attackers working from patch-to-exploit timelines, so assume the clock is running.
  • Rank by exposure first, CVSS second. A 9.8 on an internet-facing EBS tier outranks a 9.9 that needs a database session.
  • The patch is only half the work. This week’s incidents point at your third-party scripts, update channels and exposed secrets, which the CSPU doesn’t touch.

What happened this week, and why it matters to an Oracle estate

Several incidents this week matter to anyone running EBS, SOA or WebLogic. Some come from aggregator trackers, so I’ve flagged where details are unverified.

IncidentWhat happenedLesson for your estate
Optimizely supply-chain attack (Nov 16)Researchers say malicious code served through the Optimizely website-testing platform reached 178 organizations, letting criminals skim payment card numbers and crypto payments from visitors. Report URI said nearly all of the infected scripts it was tracking had been cleaned by that evening. Optimizely said it blocked the vulnerable path and deployed a fix, and Equifax said it found no evidence its own systems were compromised.The victims’ own systems weren’t breached. A trusted third-party script was. If iStore, iSupplier or any customer-facing EBS page loads external JavaScript, you have the same exposure.
cPGuard update channel hijack (Nov 2)Attackers broke into the update system of a security tool on November 2 and pushed a malicious script to customers’ Linux servers that installed a backdoor and collected password hashes.Your patch pipeline and agent update channels are an attack path. Verify signatures and checksums on everything, including Oracle media.
AI-agent-driven intrusionsGitBook confirmed a targeted attack accessed some infrastructure credentials, though the AI-agent detail is unverified. A threat-intel report claims an autonomous agent registered at least 1,438 accounts on a scanning service to hunt exposed .env files and validate stolen AI API keys. That second claim comes from one vendor report.Attackers are automating credential discovery at scale. Secrets in repos, .env files, integration configs and ISG service accounts are the soft target.
Edge and gateway exploitation (Nov 10–13)CISA added a Barracuda Email Security Gateway pair that chains an auth bypass with command injection to give root, plus an exploited Chrome use-after-free, to its KEV catalog on Nov 10. An UnrealIRCd command-injection flaw was added too.Attackers keep going after internet-facing infrastructure and chaining flaws. Think about chains, not single CVEs.
Third-party disclosuresA November 13 Uber disclosure has intensified concern over third-party security.Fourth-party risk is now a board topic. Major Cyberattack Targets 178 Organizations, Millions of Credit Card Users at Risk +8

The Oracle-specific backdrop

Oracle estates are a proven target, and the history sets expectations for the next 30 days:

  • The Clop extortion gang exploited an EBS zero-day (CVE-2025-61882, CVSS 9.8) against multiple victims, including US universities. One report says the group stole data from more than 100 organizations. Extortion demands ranged from millions up to $50 million in at least one case. www.bleepingcomputer.com +3
  • Oracle mitigated a PeopleSoft zero-day (CVE-2026-35273) that ShinyHunters exploited for unauthenticated remote code execution between May 27 and June 9. CISA also flagged a WebLogic flaw (CVE-2024-21182) as actively exploited two years after it was patched. Bleeping ComputerBleeping Computer
  • Another EBS flaw, CVE-2026-46817 (9.8, unauthenticated HTTP takeover), was seen exploited on honeypots before Oracle had flagged it as exploited. Bleeping Computer
  • CISA has listed 44 Oracle vulnerabilities as exploited, 13 of which were also abused by ransomware gangs. Bleeping Computer

The pattern is consistent. Attackers hit unauthenticated HTTP paths in EBS, PeopleSoft and WebLogic, steal data, then extort. Data theft, not encryption, is the main business risk. That should shape your priorities.

The architect’s read of the CSPU: think in chains

A CVSS list tells you what’s broken. An architect asks what an attacker can reach from the internet and what they can pivot to. Here is the November advisory as a chain:

Edge → web tier → application → middleware → database

  1. Edge and web tier. Oracle HTTP Server’s WebLogic proxy plug-in (curl component, CVE-2026-9545, 7.5, unauthenticated, HTTPS) sits in front of many EBS deployments. APEX and ORDS also carry CVE-2026-114849 (9.8, unauthenticated, HTTPS), and APEX has a second 9.8 (CVE-2026-114850). Many EBS shops run APEX or ORDS for side applications, and they’re easy to overlook in an EBS-focused patch plan. CVE-2026-114849 also credits several researchers working with Trend Micro’s Zero Day Initiative. That is my inference, not Oracle’s, but multiple independent reporters suggest more people have looked at this bug.
  2. Application tier. The four unauthenticated 9.8s in EBS (Attachments, AOL Core, OAF REST Services, Universal Work Queue media integration), plus Mobile Field Service Admin UI (CVE-2026-101691, 9.3).
  3. Middleware. WebLogic (TopLink at 10.0 and 9.8, T3/IIOP and SOAP at 9.8), Oracle Platform Security for Java (10.0) and Access Manager (three 10.0s). A foothold here reaches EBS, SOA and everything else on the domain.
  4. Management plane. Enterprise Manager’s Agent Next Gen has a Jetty issue (CVE-2026-10050, 9.1, unauthenticated). If OEM manages your EBS and database, it is a lateral-movement path.
  5. Database. Three 9.9 flaws need only Create Session. After an application-tier compromise, that is easy to obtain.

The attacker doesn’t care which product family a CVE sits in, so your patch groups shouldn’t either. Treat the EBS + OHS + WebLogic + OAM + database + OEM group as one release.

EBS deep dive

Tier-0 EBS items (unauthenticated, network):

CVEModuleScore
CVE-2026-114880Document Mgmt & Collaboration, Attachments9.8
CVE-2026-114924Application Object Library, Core9.8
CVE-2026-114925Applications Framework, REST Services9.8
CVE-2026-114863Universal Work Queue, Media Integration9.8
CVE-2026-101691Mobile Field Service, Admin UI9.3 (scope changed)
CVE-2026-114881Document Mgmt & Collaboration, Attachments9.1
CVE-2026-114876iStore, Shopping Cart8.6 (scope changed)

What the module spread tells you:

  • Customer- and supplier-facing modules (iStore, iProcurement, Sourcing, Mobile Field Service) carry several unauthenticated issues. If you publish any of these through an external tier, they go to the front of the queue.
  • Purchasing (9), iProcurement (7) and Sales Online (6) are mostly authenticated flaws at 7.5–8.8. These matter after an account takeover, which is the usual route in once a credential is stolen.
  • Mobile Field Service has 11 entries, mostly in the Admin UI. If you don’t use this module, confirm it isn’t deployed and consider whether it should be.
  • CVE-2026-114918 (Concurrent Processing, BI Publisher Integration, 8.7) also addresses CVE-2025-61882, the Clop zero-day. Oracle’s matrix lists it as needing high privileges, which differs from how the 2025 flaw was exploited, so don’t read too much into the scoring. Verify the patch is installed, not just scheduled.
  • BI Publisher has three new items (CVE-2026-114797 and CVE-2026-114789 at 6.1, CVE-2026-114810 at 5.5 in the EBS-XDO component). Most EBS shops run BIP, so include it.

Also check Java. The advisory lists 11 Java SE issues, all remotely exploitable, including server-side API paths. EBS and WebLogic depend on the JDK, so schedule JDK updates with the same release.

SOA and integration deep dive

Integration is the layer most likely to break during patching and the least tested afterward.

EBS Integrated SOA Gateway (ISG). The three new ISG patches (CVE-2026-114903, -114901, -114904) are modest in score. But ISG publishes business logic as web services, and service accounts there are often over-privileged and rarely rotated. After this week’s reports of automated credential harvesting, audit those accounts at the same time as patching.

Service Bus and SOA Suite. Service Bus has an RMI flaw at 9.6 (CVE-2026-91561) and an HTTP flaw at 8.8 (CVE-2026-91566), both needing low privileges. Any proxy service that accepts external partner traffic is a place to look first. SOA Suite and BPM carry third-party component issues (Bouncy Castle, Apache HttpClient, HttpComponents), plus a Human Workflow flaw.

Managed File Transfer. CVE-2026-8763 (9.1, HTTPS, unauthenticated, Bouncy Castle) and CVE-2026-114827 (8.8). If MFT moves payroll, banking or supplier files, this is a data-theft target, which is the Clop playbook.

WebLogic. Several issues are rated 10.0 or 9.8 without authentication. Patch them everywhere, including non-production domains that share credentials or network paths with production.

Regression risk. Patched middleware can change TLS defaults, cipher support, serialization filters and library versions. Build an integration regression pack (one test per partner interface, with message-level assertions) before you patch. Without it, you won’t find out what broke until a partner calls.

The product owner’s view: turn the advisory into a backlog

Treat the CSPU as a product release with business-facing consequences.

1. Map patches to business capabilities, not just components.
Which processes depend on the affected modules? Procure-to-pay (Purchasing, iProcurement, Sourcing), order-to-cash (iStore, Order Management), field operations (Mobile Field Service), and supplier and partner integration (ISG, MFT, Service Bus). Rank capabilities by revenue and regulatory impact. That gives stakeholders a way to approve downtime they can understand.

2. Mind the calendar.
Today is 17 November. Thanksgiving week, Black Friday and Cyber Monday load iStore and order flows. Year-end close is approaching, and many finance teams freeze changes in December. Oracle’s next releases are 15 December 2026 (CSPU) and 19 January 2027 (CPU). A freeze that skips the November patch means carrying it through two more releases. Get the freeze exceptions approved now, not mid-December.

3. Slice the work by risk tier.

TierScopeTarget
0Internet-facing and unauthenticated: EBS 9.8s, iStore, Mobile Field Service, OHS/WebLogic/OAM/OPSS 9.8–10.0, APEX/ORDS 9.8s48–72 hours, or compensating controls with a dated expiry
1Authenticated 7.5+ in EBS and Service Bus, database 9.9s, ISG, MFT, OEM, Java7–10 days
2Remaining medium and low items, third-party components with VEX justificationsNext planned maintenance window

4. Define “done” so it can be audited.
For each Tier-0 item: patch applied in production, version verified by query, integration regression passed, evidence attached to the change ticket, and exposure re-scanned. “Scheduled” doesn’t count.

5. Communicate in business terms.
Executives don’t need CVE numbers. They need: “These patches protect against the data-theft pattern that hit EBS customers in 2025. Here is the downtime and the risk if we defer.”

The chief architect’s view: patch, then reduce what needs patching

Patch cadence alone won’t keep up. These are the structural moves that make the next CSPU less urgent:

  1. Shrink the external surface. Publish only the specific URLs external users need through a reverse proxy or the EBS external-tier and URL-allowlist approach. Block admin UIs, REST and ISG endpoints, and diagnostic paths from the internet. Many of this month’s unauthenticated EBS items sit in paths that most customers don’t need externally.
  2. Segment the stack. The edge, EBS application tier, SOA and WebLogic domains, MFT and the database should have restricted east-west paths. Block T3, IIOP and RMI from anywhere that doesn’t need them.
  3. Control egress. Data-theft extortion needs an exit path. Restrict outbound traffic from application and middleware servers to known destinations, and alert on large transfers from attachment and file stores.
  4. Fix secrets. Move service-account credentials out of config files and repos into a vault, and rotate them on a schedule. Given this week’s automated key-hunting reports, assume exposed secrets will be found.
  5. Control third-party code. For any customer-facing page, inventory every external script, pin versions with Subresource Integrity where possible, and apply a strict Content-Security-Policy. This is the direct lesson from Optimizely.
  6. Secure the patch pipeline. Verify checksums on patch media, restrict who can stage patches, and monitor agent update channels.
  7. Reduce data at rest. Attachments, archived transactions and old supplier records are what extortionists steal. Purge and archive aggressively, and encrypt what remains.
  8. Detect, not just prevent. Log WebLogic, OHS and EBS access, alert on unusual requests to admin and REST paths, and watch for new concurrent-request definitions and unexpected OS-level processes on application servers. You need to be able to answer “did anyone exploit this before we patched?”
  9. Invest in rapid patching. Use online patching, automated pre- and post-checks, and a standing monthly cadence so a critical patch is routine and not a project.

The GRC angle

Auditors will treat this cycle as a control test. Beyond the basics (SLAs, evidence, risk acceptances, supported-version status, earlier Security Alerts for CVE-2026-91403, -114820, -109618 and -109619), add these:

  • Map to your frameworks. Document how the work satisfies vulnerability-management and change-management controls (SOX ITGCs, ISO 27001, NIST CSF, PCI DSS if iStore handles cards). One evidence package can serve several.
  • Track a few KRIs: time to patch Tier-0 items, percent of external endpoints on the allowlist, number of open risk acceptances past expiry, and count of service accounts with stale credentials.
  • Run a “were we hit?” review. For internet-facing EBS and WebLogic, review logs for the period before patching. Patching closes the door but doesn’t tell you whether someone was already inside.
  • Extend vendor-risk management to fourth parties. Optimizely-style incidents show up as “a vendor of my vendor.” Ask critical suppliers how they manage third-party scripts and update channels.
  • Rehearse the extortion scenario. Decide in advance who talks to legal, regulators, insurers and the board if you receive an extortion email, and who decides whether to engage. Don’t write the playbook during the incident.
  • Handle compensating controls honestly. Oracle notes that blocking protocols or removing privileges can break functionality and isn’t a long-term fix. Log each as a time-bound exception with an owner and expiry.
  • Use VEX statements properly. Oracle lists some third-party flaws as non-exploitable (for example, “vulnerable code not in execute path”). Record these as justified exceptions so auditors see a decision, not an omission.

Sunday, October 4, 2026

Beyond the Create Button: Architecting Oracle Autonomous Database on Microsoft Azure into a Live Data Pipeline

1. Create blade: pick Transaction Processing:

 




Why OLTP: the app writes small, concurrent rows (customer profiles, orders, sessions). Transaction Processing tunes for high concurrency and short statements; Data Warehouse favors large scans. Pick it first because it cannot be switched in place later.


2. Database configuration in depth, with auto scaling on







SettingValueReasoning
Database version19cLong-term-support release; broadest driver and tool compatibility.
ECPU (base)2Start small; you pay for base ECPUs continuously, so size for the quiet period.
Compute auto scalingOnLets the database burst to roughly 3x base on demand and fall back; you are billed for actual use above base. Covers peaks without a resize event.
Storage1 TB + auto scalingAuto scaling prevents a full-disk stall on growth. Set a budget alert since storage can only grow elastically.
Backup retention60 daysAutomatic backups are managed by Oracle and billed separately from database storage. Older backups are deleted past the window.
Admin credentialsADMINUse a long unique password stored in Key Vault; create named app users and avoid daily ADMIN use.
LicenseLicense includedSwitch to BYOL only if your Oracle agreement allows it.
NetworkingPrivate endpoint on a delegated subnet, NSG attachedNo public access; mTLS optional (see Screen 4).


Equivalent CLI



3. Deployment in progress:


Takeaway: the Azure resource appears fast, but the database lifecycle completes in OCI. Wait for the OCI state to read Available before connecting.


4. OCI console: nanubalu is Available:



Takeaway: every choice from Screen 2 is confirmable here. Buttons across the top (Database actions, Database connection, Performance hub, Manage resource allocation) are the daily-ops entry points. Verify from SQL:


SELECT name, open_mode, database_role FROM v$database;

SELECT * FROM v$version;

SELECT * FROM v$services;     -- HIGH / MEDIUM / LOW / TP / TPURGENT

SELECT value FROM v$parameter WHERE name = 'cpu_count';


5. MongoDB Compass: saved queries on the source


Takeaway: Oracle will hold the transactional and relational side of the same application.

6. Querying movieCollection:





Takeaway: nested arrays (cast, crew, awards) are why a document model suits the catalog while customer demographics stay relational.

7. Data Factory: factory resources:


Takeaway: one pipeline, two datasets: a CSV source in ADLS and an Oracle sink pointing at nanubalu.


8. Source dataset on the bronze layer:



Takeaway: "bronze" signals a raw landing zone; keep it immutable and let the pipeline shape data on the way into Oracle.


9. Pipeline run: Copy data:



Takeaway: add a trigger (schedule or storage event) once the manual run succeeds, and alert on failure in Azure Monitor.


10. Verify in SQL:


begin
  dbms_cloud_repo.install_file(
    repo          => get_repo,
    file_path     => 'tables/demographics.sql',
    branch_name   => 'main',
    stop_on_error => false);
end;
/
select * from demographics;

Query result: fetched 200 rows in 0.238 seconds

CUST_IDLAST_NAMEFIRST_NAMEEMAILAGECOMMUTECREDITEDUCATION
12386074123426LessThanHS
1239220524019High School
12420774417492Bachelors

Takeaway: install the DDL from a Git repo with DBMS_CLOUD_REPO, then prove the load. Add these checks and a masking rule so personal data never reaches lower environments:

SELECT COUNT(*) FROM demographics;                       -- expect 200

SELECT education, ROUND(AVG(credit_balance),0) avg_credit, COUNT(*) n

FROM   demographics GROUP BY education ORDER BY n DESC;


-- Mask email for a read-only analyst role

BEGIN

  DBMS_REDACT.ADD_POLICY(

    object_schema => 'ADMIN', object_name => 'DEMOGRAPHICS',

    column_name   => 'EMAIL', policy_name => 'MASK_EMAIL',

    function_type => DBMS_REDACT.PARTIAL,

    function_parameters => 'VVVVVVVVVVVVVVVVVVVV,VVVVVVVVVVVVVVVVVVVV,*,1,20',

    expression    => 'SYS_CONTEXT(''USERENV'',''SESSION_USER'') != ''ADMIN''');

END;

/



Auto scaling: how to prove it works:


-- Current vs base CPU allocation

SELECT name, value FROM v$parameter WHERE name IN ('cpu_count','resource_manager_plan');


-- Peak usage in the last hour from Active Session History

SELECT TRUNC(sample_time,'MI') minute, COUNT(*)/60 avg_active_sessions

FROM   v$active_session_history

WHERE  sample_time > SYSTIMESTAMP - INTERVAL '1' HOUR

GROUP  BY TRUNC(sample_time,'MI') ORDER BY 1;

Watch the Performance hub and the OCI metrics CpuUtilization and StorageUtilization during a load test; alert at 80 percent so you notice sustained scale-up before the bill does.

Architecture at a glance

Control plane is split by design: Azure owns the resource, networking, RBAC, tags and invoice; OCI owns database lifecycle (scale, backup, clone, Data Guard, Performance Hub). Write that into your RACI before go-live.

Sizing and auto-scaling mechanics

TopicWhat to know
ECPU modelECPU is the current billing unit for Autonomous (replacing OCPU). It is a hardware-agnostic compute unit, so do not carry OCPU counts over one-to-one; benchmark.
Compute auto scalingAllows up to 3x the base ECPU. Base is billed continuously; usage above base is billed for the time it is actually used. Size base near your steady p50 and let burst absorb p95 to p99.
Storage auto scalingLets storage grow automatically (up to a multiple of the reserved size) so a load never fails on a full tablespace. It protects availability, not your budget, so add a cost alert.
Memory and IOMemory and IOPS scale with ECPU. Scaling compute online needs no downtime.
Tuning ruleIf the database sits at 2x to 3x for hours every day, raise the base. Burst is for spikes, and sustained use at burst is cheaper as a larger base.
Idle cost controlUse the auto start/stop schedule on non-production copies (it is Disabled on this instance, which suits prod).

Connection services: pick the right consumer group

ServiceUse it forBehavior
TPURGENTTime-critical OLTPHighest priority; parallelism only by hint.
TPDefault app trafficNo automatic parallelism; high concurrency. Use this for the application.
HIGHReports, ETL, adminLargest share of resources, parallel queries, fewer concurrent statements.
MEDIUMMixed jobsBalanced share with limited parallelism.
LOWBackground workLowest priority, highest concurrency, serial execution.


Point the Data Factory sink at HIGH or MEDIUM so bulk loads do not starve the app on TP.
-- Which service is this session using?
SELECT SYS_CONTEXT('USERENV','SERVICE_NAME') svc,
       SYS_CONTEXT('USERENV','SESSION_USER')  usr FROM dual;

-- Who is connected, by service, right now?
SELECT service_name, COUNT(*) sessions
FROM   v$session WHERE type = 'USER' GROUP BY service_name;
Connectivity and transport security:

# python-oracledb, thin mode, TLS, no wallet (mTLS not required)
import oracledb
dsn = ("(description=(retry_count=3)(retry_delay=2)"
       "(address=(protocol=tcps)(port=1521)(host=nanubalu-db.nwf.internal))"
       "(connect_data=(service_name=<your_service>_tp.adb.oraclecloud.com))"
       "(security=(ssl_server_dn_match=yes)))")
pool = oracledb.create_pool(user="APP_USER", password=PW, dsn=dsn,
                            min=2, max=20, increment=2)
ItemGuidance
Access typePrivate endpoint on a delegated subnet; no public IP. Resolve the endpoint through an Azure Private DNS zone and give apps a CNAME, never a hardcoded host.
Ports1522 for mutual TLS (wallet required), 1521 for one-way TLS when mTLS is "Not required" (as on this instance). Allow only the app and integration subnets in the NSG.
mTLSLeave it off only if the network is private and an NSG/ACL restricts sources. If clients are untrusted or cross-network, require mTLS and rotate the wallet on a schedule.
Network ACLAdd an access control list in addition to the NSG: two independent allow-lists beat one.

Resilience: backups are not disaster recovery

CapabilityState on nanubaluExpert note
Automatic backupsOn, 60 daysEnables point-in-time restore inside the window. Billed separately from storage.
Long-term backupsNot scheduledUse for compliance retention beyond 60 days (up to years).
Local standby (Autonomous Data Guard)Backup-based onlyEnable for a faster, near-zero-data-loss failover within the region.
Cross-regionNot enabledEnable a cross-region standby for regional outages; plan DNS failover with a short TTL.
Restore testNot yet doneClone from a backup quarterly and run your validation queries. An untested backup is a hope.

Write the targets down: for example RPO 15 minutes and RTO 1 hour, then choose backup-only, local standby or cross-region standby to meet them, not the other way round.

Performance and tuning like an OCI admin

  • Auto indexing and automatic statistics are on by default in Autonomous; review them, do not fight them.
  • Performance Hub and Real-Time SQL Monitoring for live diagnosis; Operations Insights for capacity trends.
  • Resilient clients: use connection pools with retry in the connect descriptor and configure Application Continuity or Transparent Application Continuity so maintenance and scaling events are invisible to users.
-- Auto indexing: mode and what it did in the last day
SELECT parameter_name, parameter_value
FROM   dba_auto_index_config WHERE parameter_name IN ('AUTO_INDEX_MODE','AUTO_INDEX_SCHEMA');

SELECT DBMS_AUTO_INDEX.REPORT_ACTIVITY(SYSTIMESTAMP-1, SYSTIMESTAMP, 'TEXT') FROM dual;

-- Top SQL by elapsed time per execution
SELECT sql_id, executions, ROUND(elapsed_time/1e6,1) total_s,
       ROUND(elapsed_time/NULLIF(executions,0)/1e3,1) ms_per_exec,
       SUBSTR(sql_text,1,80) txt
FROM   v$sql WHERE executions > 0
ORDER  BY elapsed_time DESC FETCH FIRST 10 ROWS ONLY;

Infrastructure as code

Provision from code so the portal clicks in Screens 1 and 2 are repeatable. Illustrative Terraform (check current azurerm resource and argument names before use):

resource "azurerm_oracle_autonomous_database" "nanubalu" {
  name                             = "nanubalu"
  resource_group_name              = "MovieStream"
  location                         = "eastus"
  subnet_id                        = azurerm_subnet.oradata.id
  virtual_network_id               = azurerm_virtual_network.data.id
  display_name                     = "nanubalu"
  db_workload                      = "OLTP"        # Transaction Processing
  db_version                       = "19c"
  compute_model                    = "ECPU"
  compute_count                    = 2
  auto_scaling_enabled             = true          # compute burst
  auto_scaling_for_storage_enabled = true
  data_storage_size_in_tbs         = 1
  backup_retention_period_in_days  = 60
  license_model                    = "LicenseIncluded"
  mtls_connection_required         = false
  admin_password                   = var.adb_admin_password   # from Key Vault
  tags = { workload = "moviestream", env = "prod", owner = "data-platform" }
}

Hardening the Data Factory pipeline

  • Run the copy through a self-hosted integration runtime in the same VNet so traffic stays on the private endpoint.
  • Make loads idempotent: land in a staging table, then MERGE. Re-running a failed pipeline must not duplicate rows.
  • Tune the copy activity: batch size, parallel copies and the sink service (MEDIUM/HIGH) to match the ECPU you have.
  • Alternative: let the database pull from object storage with DBMS_CLOUD.COPY_DATA, which is often faster for large files.
-- Idempotent upsert from staging
MERGE INTO demographics d
USING demographics_stg s ON (d.cust_id = s.cust_id)
WHEN MATCHED THEN UPDATE SET d.age = s.age, d.credit_balance = s.credit_balance,
                             d.education = s.education
WHEN NOT MATCHED THEN INSERT (cust_id,last_name,first_name,email,age,commute_distance,credit_balance,education)
  VALUES (s.cust_id,s.last_name,s.first_name,s.email,s.age,s.commute_distance,s.credit_balance,s.education);

-- Post-load reconciliation
SELECT (SELECT COUNT(*) FROM demographics_stg) staged,
       (SELECT COUNT(*) FROM demographics)     loaded FROM dual;

FinOps for the architect

  • Cost drivers: base ECPU hours, burst ECPU time, storage, backup storage and (if used) cross-region standby.
  • Tag every resource (workload, env, owner, costcenter) so Azure Cost Management can show Oracle beside AKS and Data Factory.
  • Review the base ECPU monthly using peak-to-base ratios; right-size before buying more.
  • Confirm with both vendors how Marketplace spend counts toward your Azure commitment.

Troubleshooting cheat sheet

SymptomUsual causeFirst check
ORA-12541 / ORA-12170 timeoutNSG, ACL, DNS or wrong portResolve the host from the client subnet; test port 1521 or 1522.
ORA-28759 / wallet errorsmTLS required but no wallet, or expired walletMatch port to the mTLS setting; re-download the wallet.
ORA-01017Wrong or expired password, or the wrong userCheck dba_users.account_status; unlock and reset.
Copy slow or throttledSink on a low-priority service, or base ECPU too smallSwitch to MEDIUM/HIGH; watch Performance Hub for CPU and IO waits.
Unexpected billSustained burst, backup growth or storage auto-scaleCost analysis by tag; compare base to peak ECPU.

Oracle’s November Patch Storm: What EBS, SOA and WebLogic Teams Must Fix First

  Evidence Pack Share of new patches that are remotely exploitable without authentication Oracle APEX 2 / 2 Java SE 11 / 11 Fusion Middlewar...