Skip to content

Commit dea80a5

Browse files
committed
more on UGI; Add TODO markers
1 parent 9b501e1 commit dea80a5

File tree

3 files changed

+21
-6
lines changed

3 files changed

+21
-6
lines changed

sections/hdfs.md

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -31,6 +31,8 @@ code.
3131

3232
## HDFS Namenode
3333

34+
### TODO
35+
3436
1. Namenode reads in a keytab and initializes itself from there (i.e. no need to `kinit`; ticket
3537
renewal handed by `UGI`).
3638
1. In a secure cluster, Web HDFS requires SPNEGO
@@ -41,8 +43,11 @@ to access the HDFS browser. This is a point of contention: its implicit from the
4143
as "dr who".
4244

4345

46+
4447
## Datanodes
4548

49+
### TODO
50+
4651
## HDFS Client interaction
4752

4853
1. Client asks NN for access to a path, identifying via KST or DT.

sections/ugi.md

Lines changed: 8 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -14,29 +14,32 @@
1414

1515
# UGI
1616

17+
> From the pictures I turned to the bulky, closely written letter itself; and for the next three hours was immersed in a gulf of unutterable horror. Where Akeley had given only outlines before, he now entered into minute details; presenting long transcripts of words overheard in the woods at night, long accounts of monstrous pinkish forms spied in thickets at twilight on the hills, and a terrible cosmic narrative derived from the application of profound and varied scholarship to the endless bygone discourses of the mad self-styled spy who had killed himself.
18+
19+
> HP Lovecraft [The Whisperer in Darkness](http://www.hplovecraft.com/writings/texts/fiction/wid.aspx), 1931
1720
1821
If there is one class guaranteed to strike fear into anyone with experience in Hadoop+Kerberos code it is `UserGroupInformation`, abbreviated to "UGI"
1922

23+
Nobody says `UserGroupInformation` out loud; it is the *him which must not be named* of the stack
2024

2125
## What does UGI do?
2226

2327
Here sre some of the things it can do
2428

2529
1. Handles the initial login process, using any environmental `kinit`-ed tokens or a keytab.
2630
1. Spawn off a thread to renew the TGT
27-
1. Provides an operation for-on demand verification/re-init of kerberos tickets details before issuing a request.
28-
29-
31+
1. Support an operation for-on demand verification/re-init of kerberos tickets details before issuing a request.
32+
1. Appear in stack traces which warn the viewer of security related trouble.
3033

3134

32-
## UGI strengths
35+
## UGI Strengths
3336

3437
* It's one place for almost all Kerberos/User authentication to live.
3538
* Being fairly widely used, once you've learned it, your knowledge works through
3639
the entire Hadoop stack.
3740

3841

39-
## UGI troublespots
42+
## UGI Troublespots
4043

4144
* It's a singleton. Don't expect to have one "real user" per process.
4245
This does sort of makes sense. Even a single service has its "service" identity; as the

sections/yarn.md

Lines changed: 8 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,14 @@ YARN applications are somewhere where Hadoop authentication becomes some of its
1919
Anyone writing a YARN application will encounter Hadoop security, and will end up spending
2020
time debugging the problems. This is "the price of security".
2121

22-
# Securing REST APIs
22+
## YARN Service security
23+
24+
### TODO
25+
26+
## Securing YARN Application REST APIs
27+
28+
29+
### TODO
2330

2431
## Strategies for token renewal on YARN services
2532

0 commit comments

Comments
 (0)